Method and device for constructing digital power system based on multifunctional intelligent agents

Abstract

The present application discloses a method and device for constructing a digital power system based on multifunctional intelligent agents. The method comprises: respectively extracting feature production element data and control production element data from power data of a power plant side and power data of a power grid side; constructing a target data vector on the basis of a power function requirement, and determining key control production element data associated with the power function requirement; using an artificial intelligence algorithm to learn a mapping function between the target data vector and the key control production element data, constructing a functional operator using the mapping function as a core, and constructing an intelligent agent on the basis of the functional operator; and adding intelligent agents corresponding to a plurality of different power function requirements into a digital power system as power function implementation units. The system can realize data-driven diversified and digital power functions, and implement more accurate capturing of the relationship between data, thereby forming high-value data assets in the power system.

Description

Method and apparatus for constructing a digital power system based on multifunctional intelligent agents Technical Field

[0001] This application relates to the field of power system technology, and in particular to a method and apparatus for constructing a digital power system based on a multifunctional intelligent agent. Background Technology

[0002] With the advancement of industrialization and the progress and development of science and technology, international cooperation in the power energy sector is gradually strengthening. A new global electricity market is emerging: new energy sources are becoming the main force, while fossil fuels are playing a supporting role. In this transformation, new players such as new energy storage, virtual power plants, and smart microgrids are emerging, injecting new momentum into the power system. At the same time, the entire power system industry chain also faces numerous challenges.

[0003] The development and advancement of the global electricity trading market hinges on an accurate understanding and keen control of the power system. Therefore, building an intelligent and digital power system is urgently needed. Currently, the construction of the power system faces challenges of complexity and variability. Complexity lies in the diverse types of power plants connected to the grid, posing significant challenges to grid dispatch. Variation is reflected in the uncertainties of some renewable energy generation methods (such as photovoltaics and hydropower), which also greatly affects the reliability and stability of the power supply system. Therefore, there is an urgent need to integrate emerging fields such as new types of electricity and intelligent hydropower to construct an intelligent and digital power system that supports domestic electricity market reform and global electricity trading. Summary of the Invention

[0004] This application provides a method and apparatus for constructing a digital power system based on a multifunctional intelligent agent. The purpose is to construct an intelligent agent with rich functions by extracting production factor data from the power plant side and the power grid side, and to assist the power system in realizing diversified and digital power functions based on data driving through the intelligent agent with diverse functions.

[0005] The first aspect of this application provides a method for constructing a digital power system based on a multifunctional intelligent agent, the method comprising:

[0006] From the power data on the power plant side, extract characteristic production factor data corresponding to the characteristic production factor group; and from the power data on the power grid side, extract control production factor data corresponding to the control production factor group; the characteristic production factor group includes characteristic production factors at multiple levels; the control production factor group includes control production factors from multiple aspects.

[0007] Based on the power function requirements of the digital power system, a target data vector is constructed using the extracted characteristic production factor data, and key control production factor data associated with the power function requirements are determined from the extracted control production factor data.

[0008] Based on the target data vector and the key control production factor data, an artificial intelligence algorithm is used to learn the mapping function between the target data vector and the key control production factor data, a functional operator with the mapping function as its core is constructed, and an intelligent agent is constructed based on the functional operator.

[0009] Various intelligent agents corresponding to different power function requirements are added to the digital power system, and each intelligent agent serves as a power function implementation unit in the digital power system.

[0010] A second aspect of this application provides a device for constructing a digital power system based on a multifunctional intelligent agent, the device comprising:

[0011] The data extraction module is used to extract characteristic production factor data corresponding to the characteristic production factor group from the power data on the power plant side; and to extract control production factor data corresponding to the control production factor group from the power data on the power grid side; the characteristic production factor group includes characteristic production factors at multiple levels; the control production factor group includes control production factors from multiple aspects.

[0012] The vector construction module is used to construct a target data vector based on the power function requirements of the digital power system using the extracted characteristic production factor data.

[0013] The data determination module is used to determine the key control production element data associated with the power function requirements from the extracted control production element data;

[0014] The operator construction module is used to learn the mapping function between the target data vector and the key control production factor data using artificial intelligence algorithms based on the target data vector and the key control production factor data, construct a functional operator with the mapping function as the kernel, and construct an intelligent agent based on the functional operator.

[0015] The intelligent agent adding module is used to add intelligent agents corresponding to various different power function requirements to the digital power system, and to use each intelligent agent as a power function implementation unit in the digital power system.

[0016] Optionally, in one implementation of the first and second aspects, the group of characteristic production factors includes primary, secondary, and tertiary characteristic production factors corresponding to various different power generation types; wherein, the primary characteristic production factors are natural characteristic production factors, the secondary characteristic production factors are single-system characteristic production factors, and the tertiary characteristic production factors are plant system characteristic production factors; the natural characteristic production factors are production factors that directly reflect natural characteristics; the single-system characteristic production factors are production factors involving individual systems in the power system; and the plant system characteristic production factors are production factors involving the overall system in the power system.

[0017] Optionally, in one implementation of the first and second aspects, the control of the production factors in the plurality of aspects includes: electricity price, inertia and frequency, active and reactive power, carbon emissions from electricity, safety and stability, and grid connection and disconnection;

[0018] The extraction of control production factor data corresponding to the control production factor group from the power grid side includes:

[0019] From the power data on the grid side, extract electricity price data, inertia and frequency data, power data, carbon emission data, safety and stability index data, and grid connection and disconnection impact data.

[0020] Optionally, the power function requirements are: power function requirements for power trading between the grid side and the power plant side, power function requirements for realizing the grid side's dispatch control over the power plant side, or power function requirements for the grid's own regulation.

[0021] Regarding the first and second aspects, in a first possible implementation, the power function requirement for power trading between the grid side and the power plant side specifically involves calculating electricity prices, and the functional operator is an electricity price operator used to calculate electricity prices; the process by which the digital power system utilizes an intelligent agent constructed based on the electricity price operator includes:

[0022] The intelligent agent is invoked to obtain the electricity price calculation result on the power plant side through the electricity price operator;

[0023] The electricity price calculation results are verified using a financial verification model on the grid side. The financial verification model includes: a first verification condition and a second verification condition. The first verification condition is: the actual output of various types of power plants is greater than or equal to the grid power volume. The second verification condition is: the sum of the products of the actual output of various types of power plants, the grid electricity price at the corresponding time, and the power generation duration is less than or equal to the product of the grid average price and the grid power volume.

[0024] If the electricity price calculation result meets both the first verification condition and the second verification condition, then the electricity price calculation result is determined to have passed verification; if the electricity price calculation result does not meet either the first verification condition or the second verification condition, then the electricity price calculation result is determined to have failed verification.

[0025] If the electricity price calculation result is verified, then a suitable and feasible electricity price calculation function for the grid side is selected from the scheduling strategy library, and the electricity price calculation function is loaded into the electricity price operator to update the electricity price operator.

[0026] The updated electricity price operator is used to obtain the electricity price calculation results on the grid side.

[0027] Regarding the first and second aspects, in the second possible implementation, the power function requirement for power trading between the grid side and the power plant side is specifically power clearing, the functional operator is a clearing operator, and the clearing operator is used for power clearing; the process by which the digital power system utilizes an intelligent agent constructed based on the clearing operator includes:

[0028] Based on the power generation element characteristics of each associated power plant in the power grid, low-level data abstraction processing is performed to obtain the first data vector of each power generation element characteristic.

[0029] Long-term transaction electricity prices are predicted based on a preset long-term electricity price function library and each of the first data vectors, so as to establish a long-term transaction stack.

[0030] A spot trading stack is established based on the spot trading electricity prices of each of the associated power plants, and a first clearing function is determined through the long-term trading stack and the spot trading stack.

[0031] Based on the first clearing function, determine the power supply and demand balance state of the power grid when it is cleared under the first clearing function;

[0032] The first clearing function is adjusted according to the power supply and demand balance state to obtain the second clearing function, and power clearing is performed based on the second clearing function.

[0033] Regarding the first and second aspects, in the third possible implementation, the power function requirement for realizing grid-side dispatch control over the power plant side specifically refers to peak-shaving dispatch in high-frequency load change scenarios of thermal power plants. The functional operator is a thermal power peak-shaving dispatch operator, which is used to realize peak-shaving dispatch in high-frequency load change scenarios of thermal power plants. The process of the digital power system utilizing an intelligent agent constructed based on the thermal power peak-shaving dispatch operator includes:

[0034] Calculate the heat storage margin of energy storage devices in a thermal power plant; the heat storage margin includes a heat storage margin characterization value and a heat release capacity characterization value;

[0035] The system sends the correlation data between the load, performance, and cost of each thermal power unit in the thermal power plant, as well as the heat storage margin, to the grid side so that the grid side can determine the target units for expected auxiliary peak shaving and issue peak shaving dispatch instructions based on the changes in thermal power load demand, the correlation data provided by each thermal power plant, and the heat storage margin.

[0036] The system receives a peak-shaving dispatch instruction from the power grid side; the peak-shaving dispatch instruction carries the unit identifier of the target unit and peak-shaving requirement information for the target unit; the peak-shaving requirement information includes the peak-shaving load curve.

[0037] Assisted peak shaving services are performed based on the peak shaving requirement information.

[0038] Regarding the first and second aspects, in the fourth possible implementation, the power function requirement for realizing grid-side dispatch control over the power plant side specifically refers to water resource dispatching of cascade hydropower stations, the functional operator is a water resource dispatching operator, and the water resource dispatching operator is used to realize water resource dispatching of cascade hydropower stations; the process of the digital power system utilizing an intelligent agent constructed based on the water resource dispatching operator includes:

[0039] Obtain water resource data for each hydropower station in the cascade hydropower station system at the current moment;

[0040] For each of the aforementioned hydropower stations, based on the water resource data of that hydropower station, it is determined whether the hydropower station is in a safe operating state;

[0041] When it is determined that all the hydropower stations are in the safe operating state, it is determined whether to carry out water resource scheduling based on the water resource data of each hydropower station, the preset electricity price scheme and the target historical data; the target historical data includes the average power generation water consumption and the average power plant revenue corresponding to the month in which the current moment is located.

[0042] If water resource scheduling is determined, a water resource scheduling scheme is determined based on the water resource data and water resource scheduling model of each hydropower station; the water resource scheduling model is a pre-trained model used to output the water resource scheduling scheme; the water resource scheduling scheme includes day-ahead scheduling and seasonal scheduling.

[0043] Regarding the first and second aspects, in the fifth possible implementation, the power function requirement for realizing grid-side dispatch control over the power plant specifically refers to grid-side dispatch of the power plant in a large-scale power system with small-scale grids. The functional operator is a large-scale grid dispatch operator, which is used to realize grid-side dispatch of the power plant in the large-scale power system with small-scale grids. The process of the digital power system utilizing an intelligent agent constructed based on the large-scale grid dispatch operator includes:

[0044] Based on the digital feature extraction of the power grid and power sources of various power generation types in the power system, the emergency reserve capacity of various power generation types that should be reserved in the power system is determined and the emergency reserve capacity is configured.

[0045] Based on the output of the power sources of the various power generation types mentioned above, determine whether the current power system conforms to the characteristics of large generators and small grids;

[0046] If it is determined that the current power system conforms to the characteristics of a large generator and a small grid, then the power grid utilizes the configured emergency reserve capacity for various power generation types to schedule power sources within the power system for voltage adjustment based on a multi-round voltage adjustment scheme, and / or schedules power sources within the power system for frequency adjustment based on a multi-round frequency adjustment scheme. The multi-round voltage adjustment scheme includes: voltage adjustment methods, voltage adjustment ranges, and voltage adjustment priority information for the various power generation types; the multi-round frequency adjustment scheme includes: frequency adjustment methods, frequency adjustment ranges, and frequency adjustment priority information for the various power generation types.

[0047] Based on the stability control requirements of the frequency dynamic characteristic index of the entire power system, the expected load of the entire network and the characteristics of the generating units at each power source, the power grid generates the output curves of the generating units at each power source in the future period; the frequency dynamic characteristic index is used to numerically characterize the amount of power change of the entire network required to cause a unit frequency change in the power system.

[0048] The power grid sends generation dispatch instructions to each power source to adjust power output; the generation dispatch instructions include the output curves of the corresponding power source units.

[0049] Regarding the first and second aspects, in the sixth possible implementation, the power function requirement for grid self-regulation specifically refers to regulating the grid structure, the functional operator is a grid structure regulation operator, and the grid structure regulation operator is used to regulate the grid structure of the power grid; the process of the digital power system utilizing an intelligent agent constructed based on the grid structure regulation operator includes:

[0050] Get the current system inertia corresponding to the current mesh structure;

[0051] The frequency change rate is calculated based on the current system inertia and the difference between the power demand and the actual power supply.

[0052] If the frequency change rate is greater than a preset first threshold, then the target mesh structure and the target setpoint corresponding to the target mesh structure are determined based on a pre-established deep learning model; the frequency change rate of the target mesh structure is less than or equal to the first threshold.

[0053] Based on the target setpoint, modify the current setpoint applied in the current grid structure, and adjust the current grid structure of the power grid to the target grid structure.

[0054] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0055] In this technical solution, characteristic production factor data corresponding to characteristic production factor groups and control production factor data corresponding to control production factor groups are extracted from power data from both the power plant and grid sides. This achieves coordination between characteristic production factors on the power plant side and control production factors on the grid side. Based on the extracted relevant production factor data, artificial intelligence technology is further utilized to construct intelligent agents with specific power functions. Sufficient preparations are made for the application of this technology: based on the power function requirements of the digital power system, a target data vector is constructed using the extracted characteristic production factor data, and key control production factor data associated with power function requirements is determined from the extracted control production factor data. Based on the target data vector and key control production factor data, an artificial intelligence algorithm is used to learn the mapping function between the target data vector and the key control production factor data, constructing a functional operator with the mapping function as its core, and constructing an intelligent agent based on the functional operator. Intelligent agents corresponding to various different power function requirements are added to the digital power system, with each intelligent agent serving as a power function implementation unit within the digital power system. Thus, the digital power system possesses intelligent agents with diverse functions, enabling the system to realize diverse, digital power functions based on data-driven principles.

[0056] In this technical solution, the construction of the intelligent agent relies on relevant production factor data from both the power plant and grid sides. Furthermore, the characteristic production factor group includes multiple levels of characteristic production factors, and the control production factor group includes multiple aspects of control production factors. Therefore, it effectively achieves the coordination of multiple levels of characteristic production factors and multiple aspects of control production factors. The intelligent agent constructed based on this effectively utilizes the diverse information in power data to realize the coordinated power functions of multiple production factors, such as power functions for power trading between the grid and power plants, power functions for grid-side dispatch control of power plants, and power functions for grid self-regulation. The intelligent agent can more accurately capture the connections between data, thereby obtaining more accurate and transferable data analysis or calculation results. It can also be understood that the digital power system constructed through this solution forms high-value data assets within the power system, enhancing the analytical value of power data. The constructed digital power system can effectively contribute to domestic power market reform and global power trading. Attached Figure Description

[0057] Figure 1A is an example flowchart of data-based intelligent agent construction in this application;

[0058] Figure 1B is a schematic diagram of an operator constructed using deep learning and reinforcement learning according to an embodiment of this application;

[0059] Figure 1C is a data structure diagram in the construction of a digital power system provided in an embodiment of this application;

[0060] Figure 1D is a diagram illustrating an implementation architecture for digital power correlation provided in an embodiment of this application;

[0061] Figure 1E is a flowchart of a method for constructing a digital power system based on a multifunctional intelligent agent according to an embodiment of this application;

[0062] Figure 1F is a schematic diagram of a power supply scenario provided in an embodiment of this application;

[0063] Figure 1G is a schematic diagram of a production factor division architecture provided in an embodiment of this application;

[0064] Figure 1H is a schematic diagram illustrating the relationship between an intelligent agent, an operator, a function, and production factor data provided in an embodiment of this application.

[0065] Figure 1I is a flowchart illustrating the implementation of a power analysis intelligent agent according to an embodiment of this application;

[0066] Figure 1J is a schematic diagram of the relationship between intelligent agents, operators, functions, and production factor data provided in another embodiment of this application.

[0067] Figure 2A is a flowchart of the grid side based on the electricity price operator;

[0068] Figure 2B is a flowchart of the power plant side based on the electricity price operator;

[0069] Figure 2C shows the signaling diagram between the power plant side, the power grid side, and the intelligent agent.

[0070] Figure 3A is a schematic flowchart of a power grid clearing method provided in an embodiment of this application;

[0071] Figure 3B is a signaling interaction diagram of a power grid clearing method provided in an embodiment of this application;

[0072] Figure 3C is a flowchart illustrating another power grid clearing method provided in an embodiment of this application;

[0073] Figure 3D is a schematic diagram of a long-term transaction stack provided in an embodiment of this application;

[0074] Figure 3E is a flowchart illustrating another power grid clearing method provided in an embodiment of this application;

[0075] Figure 3F is a schematic diagram of parameters representing the thermal performance of a unit under full load conditions, provided by an embodiment of this application.

[0076] Figure 3G is a schematic diagram of comprehensive consideration parameters for coal prices and long-term electricity trading prices provided in an embodiment of this application;

[0077] Figure 3H is a schematic diagram of a stack structure for a spot trading stack and a long-term trading stack provided in an embodiment of this application;

[0078] Figure 3I is a flowchart illustrating a method for establishing a spot trading stack according to an embodiment of this application;

[0079] Figure 4A is a schematic diagram of power prediction based on a wind power plant provided in an embodiment of this application;

[0080] Figure 4B is a schematic diagram of a spot trading stack provided in an embodiment of this application;

[0081] Figure 4C is a schematic diagram of a power grid clearing function provided in an embodiment of this application;

[0082] Figure 4D is a flowchart illustrating another power grid clearing method provided in an embodiment of this application;

[0083] Figure 4E is a flowchart illustrating an adjustment method for the first clearing function when the target power grid is in an over-generated state, according to an embodiment of this application.

[0084] Figure 4F is a schematic diagram of calculating a second data vector using a deep learning model according to an embodiment of this application;

[0085] Figure 4G is a flowchart illustrating an adjustment method for the first clearing function when the target power grid is in an under-generated state, according to an embodiment of this application.

[0086] Figure 4H is a schematic diagram of a target power grid fitting power generation curve provided in an embodiment of this application;

[0087] Figure 5A is a schematic diagram of a power supply scenario provided in an embodiment of this application;

[0088] Figure 5B is a flowchart of a peak-shaving scheduling method for thermal power plants under high-frequency load changes provided in an embodiment of this application.

[0089] Figure 5C is a signaling interaction diagram of a peak-shaving scheduling method for a thermal power plant under a high-frequency load change scenario provided in an embodiment of this application;

[0090] Figure 5D is a flowchart illustrating the implementation of determining the target generating unit for expected auxiliary peak shaving according to an embodiment of this application.

[0091] Figure 5E is a flowchart of a fault judgment and operating condition judgment provided in an embodiment of this application;

[0092] Figure 5F is a flowchart of another peak-shaving scheduling method for thermal power plants under high-frequency load changes provided in an embodiment of this application.

[0093] Figure 6A is a flowchart of a water resource scheduling method provided in an embodiment of this application;

[0094] Figure 6B is a schematic diagram of the historical power generation water consumption of an upstream hydropower station for each month, provided in an embodiment of this application.

[0095] Figure 6C is a schematic diagram of the historical power generation water consumption of a downstream hydropower station for each month, provided in an embodiment of this application.

[0096] Figure 6D is a schematic diagram of the historical power plant revenue for each month of an upstream hydropower station provided in an embodiment of this application;

[0097] Figure 6E is a schematic diagram of the historical power plant revenue for each month of a downstream hydropower station provided in an embodiment of this application;

[0098] Figure 6F is a schematic diagram of the water assets of an upstream hydropower station for each month, provided in an embodiment of this application;

[0099] Figure 6G is a schematic diagram of the water assets of a downstream hydropower station for each month, provided in an embodiment of this application;

[0100] Figure 6H is a schematic diagram of a water resource regulation scheme for a cascade hydropower station provided in an embodiment of this application;

[0101] Figure 6I is a schematic diagram of the interaction between a power plant and a power grid provided in an embodiment of this application;

[0102] Figure 6J is a flowchart of another water resource scheduling method provided in an embodiment of this application;

[0103] Figure 7A is a schematic diagram of an application scenario of a power grid dispatching method in a large-scale power system with a small grid provided in an embodiment of this application;

[0104] Figure 7B is an example flowchart of a power dispatching method in a large-scale power system with a small grid provided in an embodiment of this application;

[0105] Figure 7C is a schematic diagram of a multi-round voltage adjustment provided in an embodiment of this application;

[0106] Figure 7D is a schematic diagram of a multi-round frequency adjustment provided in an embodiment of this application;

[0107] Figure 7E is another example flowchart of the power dispatching method in a large-scale power system with a small grid provided in the embodiments of this application;

[0108] Figure 7F is a power dispatching signaling diagram in a large-scale power system with a small grid provided in an embodiment of this application;

[0109] Figure 8A is a flowchart of a method for adjusting a power grid structure according to an embodiment of this application;

[0110] Figure 8B is a flowchart of another method for adjusting a power grid structure according to an embodiment of this application;

[0111] Figure 8C is a flowchart of a method for determining the target grid structure under the condition of sudden change in power generation, provided in an embodiment of this application.

[0112] Figure 8D is a schematic diagram of a binary tree structure provided in an embodiment of this application;

[0113] Figure 9 is an interactive diagram of a power system's power generation dispatch process provided in an embodiment of this application;

[0114] Figure 10 is a schematic diagram of the relationship between a function and an operator provided in an embodiment of this application. Detailed Implementation

[0115] With increasingly extensive international cooperation in the power energy sector, the construction of digital artificial intelligence technologies that drive power intelligent agents, the promotion of a unified global power trading market, and the formulation of top-level design rules to guide the marketization process are key needs facing the power industry. Based on practical experience in the domestic and international power industries, the inventors propose a digital power system construction scheme based on multi-agent intelligence. Taking into account the demands for cost-effectiveness across the entire industry chain in the new global power market, and the digital characteristics of the production factor data contained in the power system, this scheme transforms these complex factors (coal consumption, solar radiation, wind speed, electricity pricing mechanisms, investment costs, etc.) into data or advanced data vectors, promoting cross-industry information flow and efficient utilization, and giving data new life through AI technology.

[0116] In this application, the inventors integrate forward-thinking approaches, addressing the technological and industrial challenges across the entire industry chain. They emphasize the integration of technology and economics, the digitalization of the value chain, and the simultaneous advancement of marketization and intelligentization, aiming to provide intelligent and automated solutions to power problems in the energy sector. Combining emerging fields such as new types of power and intelligent hydropower, the technical solution provided in this application offers a method and device for constructing a digital power system based on multifunctional intelligent agents, capable of serving the global energy internet and forming digital assets. Transforming data into digital assets, from the essence of artificial intelligence, requires not only the data itself but also a profound and fundamental understanding of the power industry's operating model. With the deepening of power market reform, innovation in various technologies and mechanisms has become crucial for driving reform. This application constructs a foundational design framework that deeply integrates digitalization and artificial intelligence, based on data-driven digital power, algorithms based on digital power, and intelligent agents based on the power system. Particularly in key areas such as dispatching, settlement, and pricing, it abstracts low-, mid-, and high-order characteristics of data from different dimensions to achieve a dual leap in efficiency and intelligence. The technical concept of this application can promote the transformation and application of related technologies, thereby accelerating the process of power market reform.

[0117] The following is an introduction to some of the inventive concepts involved in the technical solution of this application:

[0118] Symbiosis: From a data perspective, symbiosis refers to the digitization of a class of comparable transactions, which, under a certain abstraction, possess certain related and interconnected properties. Symbiosis integrates information about natural production factors or the production and control of these factors into the data, making data-related associations and transmission possible. Symbiosis also allows abstract data features to exist in different operators or intelligent agents simultaneously, and characteristics can exist in multiple operators and intelligent agents. Essentially, different types of transactions may share similar characteristics despite their seemingly disparate forms; finding the essence and connotation of symbiosis allows us to identify common properties. In this application's technical solution, the inventors, based on their accumulated knowledge of power systems, use symbiosis to integrate concepts from neural networks into the power system, achieving a clever application of artificial intelligence technology in the power field.

[0119] Migration: Transferring features from one type of transaction to another related type of transaction, creating a relationship and enabling the two types of transactions to be interchangeable at the data asset level. Migration is achieved through nested and invoked agents, allowing multi-functional agents to perform specific functions without requiring extensive iterations. Migration enables most functions in the power system to be transferred from one agent to another.

[0120] Emergence: When the scale of data is large enough and the logic within an intelligent entity is abundant enough, the phenomenon of data and logic mutually regenerating and iterating is observed, marking the dividing line between intelligence and wisdom. Artificial intelligence, upon reaching the emergence stage, will generate wisdom. Reflecting on how humans used tools to change productivity, these tools were the means of production. Initially in ancient times, this progressed with the continuous expansion of input data to drive productivity, leading to the invention of production tools. Then, through the summarization and abstraction of laws by disciplines such as mathematics, physics, and chemistry, logic, formulas, and theorems were ultimately derived. In modern society, the development of information science changed the carriers of the Third Industrial Revolution, giving rise to electricity, computers, and other tools for iterating information. However, in the Fourth Industrial Revolution, artificial intelligence fundamentally upgraded computers from simple to complex, using neural networks and statistical principles to approximate production functions again, allowing for a simple yet direct approach driven by data. In the next 10 years, the development of intelligent agents and their internal logic will be equivalent to the development of production tools in the preceding society. Human understanding always deepens and becomes more profound step by step. Many phenomena that cannot be explained have internal connections and deeper theorems and concepts. However, higher-level abstractions of data exist and are easier to obtain. The key is that analyzing the relevant logic within intelligent agents requires a natural process, more like waiting for a certain opportunity.

[0121] Intelligent agents: Intelligent agents are units with one or more functions, enabling a closer integration of power systems and artificial intelligence. Intelligent agents protect the power system from external influences, much like cells, but containing smaller kinetic energy units such as operators. Intelligent agents possess unique data characteristics in different scenarios, which is significant for generating emergent functions and studying the digital essence of power systems. Intelligent agents are a core component of artificial intelligence in the power industry, analogous to thermal control logic diagrams or logic-related relay protection in the power industry during the Third Industrial Revolution. For example, the logic within the "five protections" of a switching station can be symbolically represented. The logic within intelligent agents can also be symbolically represented, allowing for a more precise and deterministic representation of key logic and control within the power system, based on big data and large models. The general logic of intelligent agents conforms to key theorems and common sense within power systems, as well as energy conversion companies and related rules. Their research objects are production characteristic elements and control production elements, such as illuminance, wind speed, water level, heat consumption, load factor, system inertia, and frequency. The interface of the intelligent agent is directly connected to the input or output layer (output function) of deep learning and reinforcement learning, and also connects to intermediate layers when necessary. In the technical solution of this application, the first half of the neural network abstracts natural production elements into calibrated and transmitted data, enhancing parallel computing capabilities. The operator functions and intelligent agents in the output of the second half reflect the ability of mathematical abstraction and the physical laws within the industry, and can be combined and reconstructed, enabling the intelligent agent to possess capabilities and functions applicable to various industries.

[0122] This application's technical solution focuses entirely on constructing algorithms based on data and information flow. After feature extraction, the data needs to be labeled, which means providing a reasonable data structure. Simultaneously, the internal logic of the operators controls the information flow. Different data structures and information flows correspond to the digitalization of power plants and power grids, and are demonstrated through different implementation examples, showing that different neural networks are better suited for the algorithms. The operators involved differ in data structure and information flow methods. For example, power grids are suitable for using binary trees for grid disconnection and grid connection, and stacks for electricity price clearing, while power plants are better suited for gradually establishing information flow methods and internal safety logic based on primary indicators from nature. Considering the applicable objects, the conditions and direction of information flow are determined, which also determines the selection of convolutional neural networks, recurrent neural networks, graph neural networks, attention networks, etc. The following sections discuss related issues: I. Digital feature extraction, II. Mathematical tools, III. Binary trees and stacks in data structures, IV. Logical core, V. Reasoning ability and information classification under the underlying logic.

[0123] I. Digital Feature Extraction

[0124] Digital characteristics are divided into power plant and power grid digital characteristics. Power grid characteristic variables are established, such as the real and imaginary parts of voltage at each node, active power flowing into each node, reactive power flowing into each node, system inertia, system frequency, and the ratio of system power change rate to system frequency change rate. Power plant characteristic variables are established, such as generator power angle, voltage of each generator, electromagnetic power of each generator, and reactive power of each generator. Based on the fact that the above production factor digital characteristics originate from first-level characteristic production factors in nature, such as illuminance, coal type and quality, and water level, digital characteristics at each level are first extracted. For example, low-order characteristic production factor digital characteristics are extracted from the power plant side, such as the seasonal fluctuation of water level and the efficiency stability of illuminance; high-order control production factor digital characteristics are extracted from the power grid side, such as the system inertia versus frequency change rate characteristics and the frequency change rate versus power change rate characteristics. The methods used also differ. For the digital features of low-level production factors, such as water level and photovoltaic illuminance, clustering or Euclidean distance methods can be used to extract the digital features. However, for the digital features of high-level control production factors, modeling or more advanced mathematical tools are required to extract the digital features.

[0125] II. Mathematical Tools

[0126] Mathematical tools, depending on the characteristics of the data and information flow of the object, can employ one of the following: convolutional neural networks, recurrent neural networks, graph neural networks, or attention networks. The characteristics of these various neural networks are as follows: 1. Convolutional Neural Networks: Regular network data and information flow towards local domains; 2. Recurrent Neural Networks: Data is input sequentially, and information flows sequentially; 3. Graph Neural Networks: Data is in a fixed graph structure, and information flows along fixed edges; 4. Attention Networks: Data is in an unordered set, and information flow is dynamically controlled by the neural network. These neural networks target the logical core: operators, or intelligent agents. Typical examples corresponding to the construction of a digital power system using multifunctional intelligent agents (operators) include: grid connection and disconnection operators (Example 6 below), hydropower seasonal dispatch operators (Example 4 below), power source and grid control operators for large and small power grids (Example 5 below), and clearing function operators (Example 2 below). The data analysis and information flow characteristics in these operators profoundly reflect the aforementioned characteristics of neural networks. By analyzing the characteristics of data and information respectively from the extraction of data features to the rules of information sequence flow (physical and equipment mechanisms), and selecting suitable neural networks, the artificial intelligence algorithms are highly compatible with the power system.

[0127] When using mathematical tools, the specific tools should be determined based on the data scale and sample size. The following section, with reference to Figure 1A, illustrates an example process for constructing an intelligent agent based on data in this application. As shown in Figure 1A, the process includes:

[0128] (1) Data collection and input.

[0129] (2) Determine the data size, whether it is a small sample or a large sample. If it is a large sample, skip step (3) and execute directly for a small sample.

[0130] (3) Data fitting: Using mathematical tools or formulas to regress the data to a certain convergence value.

[0131] (4) Data cleaning: Based on the fitted data, abnormal data are deleted or processed.

[0132] (5) Extract data features to determine whether they are production factor features or control production factor features.

[0133] (6) Intelligent agent construction: Based on the internal logic of the power industry, write out the key processes or constraints, and for the constructed intelligent agents, deploy deep learning and reinforcement learning interfaces, receive data extracted from data features, and further perform deep learning or reinforcement learning based on their features and classification.

[0134] (7) Take the results of deep learning or reinforcement learning, create an output function for the output layer, and use the backpropagation algorithm to fine-tune and correct the intermediate results of each layer.

[0135] (8) Generalize the output function into control logic, bring it into the agent, and nest the agent and its key internal data elements into the control logic of the actuator to achieve data-driven operation.

[0136] Deep learning and reinforcement learning enable compressed feature transmission to have more representative and error-correcting mechanisms. Deep learning employs specific algorithms or logic for extracting data features, making these features more regular in subsequent extractions. This facilitates the use of data characteristics to build advanced mechanisms or data structures based on data cleaning; it also minimizes information loss during data transmission while ensuring data computation and transmission speed. In reinforcement learning, mechanisms such as rewards are used to continuously verify the correctness of learned logic and assign priorities, allowing logic to emerge through continuous comparison and calibration.

[0137] Based on the process described above, the digitization in this application's technical solution is not a single digitization method or a single mathematical tool. There are differences depending on the size of the sample. For digitizing large sample data, deep learning methods (such as convolutional neural networks) are required. The measured data is compared with the sample data to form a workflow. Based on the logical flow in the intelligent agent, repeated deductions and reverse actions are used to simulate human reasoning. Convolutional algorithms can obtain features through deep learning and provide them to operator functions. Convolutional neural networks adopt an "end-to-end" approach, automatically identifying data features and possessing the functions of sharing rights, hierarchical abstraction of low-order, mid-order, and high-order data features. Small sample data, on the other hand, directly extracts data features from the data.

[0138] The concept of intelligent agents is suitable for research subjects with small sample sizes. In power systems, the choice between using large models (corresponding to large samples) and small samples depends on the characteristics of the data. Currently, weather forecasting involves a large amount of data, making large models suitable. Dispatch, grid connection, and disconnection also benefit from large models due to the large data volume. Therefore, research on intelligent agents, particularly the innovative design of operator logic within power systems using deep learning and reinforcement learning, is crucial. Because computing power involves advanced processes, GPUs, and HBM, and data transmission rates and network throughput are limited, the logic within uncertain intelligent agents must adapt very quickly to the high-speed computation of GPUs. Therefore, the choice of neural network type must adapt to the characteristics of data and logical information flow, capturing logical and numerical certainty amidst uncertain changes in natural production factors. In the stages of artificial intelligence moving towards AGI / ASI, the application of intelligent agents makes small sample data more advantageous than large sample data.

[0139] For small sample data, binary trees and stacks can be used as data structures to encapsulate data features.

[0140] III. Binary Trees and Stacks in Data Structures

[0141] Encapsulating data features using binary trees or stacks allows data features to be represented more accurately or to be easily shared.

[0142] In the embodiments described below, Embodiment 2 demonstrates an example of using a stack as a data structure. After abstracting production factors, they are placed into the above data structure to achieve parallel computing and intelligent agent logic implementation. Embodiment 2 illustrates a stack method that progressively decreases in priority from bottom to top. The last-in-first-out (LIFO) structure and sequential characteristics of a stack conform to the clearing logic of the power grid and can also achieve dynamic ordering of data. For operations such as electricity price clearing targeting high-order production factors, multiple stacks or layered processing are required if high-speed calibration data parallel computing is needed. This increases the pressure of parallel computing but reduces the time for logical judgment, making it a typical case of data-driven processing.

[0143] Example 6 illustrates the use of a binary tree as a data structure. Data structures with inherent characteristics, such as binary trees, can be used to implement grid disconnection and grid connection. Practical data tools are not limited to functions; they also include graph theory and other methods.

[0144] IV. Logical Core

[0145] This paper establishes connections between three major groups, two major production factors, and four data structures. Through a logical core relating these elements, the core links production factors, operators, and agents. After data cleaning and labeling, the data characteristics representing production factors are incorporated into operators to achieve specific functions, such as electricity price, clearing, and frequency control. Agents containing these operators also acquire these functions, enabling symbiosis and transfer. Furthermore, the paper implements the underlying logic of data-driven operations and operators, as well as the subsequent reasoning process of agents, thus enabling emergence.

[0146] I. Structured and Systematized Logical Body:

[0147] 1. Construct a characteristic production factor group centered on the power plant side, including level 1-3 indicators, namely natural indicators, single-system indicators, and plant system indicators. Internal calculations are performed using parallel computation of data vectors, as this method is fast enough and only requires feature extraction from level 1-3 indicators. This allows the data to reflect the characteristics of each part, transmit features, and perform simple analysis and statistics. See the right side of Figure 1D for an example of a digital power correlation implementation architecture.

[0148] 2. Construct a control production factor group centered on the power grid, including key power grid indicators such as electricity price, inertia and frequency, active and reactive power, carbon emissions, safety and stability, and network combination and decoupling. See the left side of Figure 1D, which shows an implementation architecture of digital power correlation. The underlying data logic is functions. Based on the system rules within the intelligent agent, reinforcement learning is used to summarize the data vector mapping function. This function calculates and expresses important internal rules of the power grid. For example, by calculating system inertia, the frequency change rate is obtained. This allows the intelligent agent to call upon the control production factors of the power grid, and under reinforcement learning, the intelligent agent processes the data to achieve power grid balance and stability. Other factors, such as electricity price, are also based on the principle of digital power correlation. Various types of data vectors are mapped to functions on the power grid side, and reinforcement learning iterates these functions to obtain the optimal solution for the control production factors on the power grid side.

[0149] 3. Construct a multifunctional intelligent agent, as shown in Examples 1 to 6. The intelligent agent is a key logical entity that combines artificial intelligence and electricity to implement specific functions.

[0150] II. Datafication and Functionalization

[0151] 4. The abstract characteristics of the data reflect primary indicators and indicators in nature. These indicators need to be abstracted using mathematical tools and then quantified, normalized, and abstracted under the condition of permissible signal loss. Then, utilizing the characteristics of parallel computing, data iteration is performed at the transmission layer. Finally, at the output layer, it is functionalized, and intelligent agents are used to embody relevant rules from nature or industry, such as rules and constraints. Operators are used as functions to implement specific functions. These operators are nested within the intelligent agents, possessing certain functions. These functions are thus symbiotic with the intelligent agents. The combination and division of intelligent agents are analogous to the combination and division of cells. In this process, operators are equivalent to proteins; their different combinations and reconstructions give proteins certain functions. The data at the transmission layer transmits protein information or nutrients. This data is generalized from functions into operators, and these data originate from the characteristic abstraction of production factors. Figure 1B is a schematic diagram of an embodiment of this application providing a method for constructing operators through deep learning and reinforcement learning.

[0152] Nature and the Second Machine Revolution established the aforementioned connection based on neural networks and data labeling. This structure and interface, under the self-attention mechanism, enabled probabilistic methods to improve the accuracy of artificial intelligence. Furthermore, the emergence of GPU parallel computing power necessitates the design of mathematical abstractions (operators) and the development of intelligent agents in the power industry and various other sectors. Featured production factors serve as the characteristic carriers of various abstract indicators. Relevant power plant grid indicators are digitized into data vectors representing their characteristics. After feature extraction, the data is fed into the operators.

[0153] 5. In data structures, numerical vectors are key to reflecting power indicators at power plants. For example, measuring the impact of fog on production starts with data vectors from each power plant (number of fog days, power affected by fog, etc.), and then uses deep learning through an intelligent agent interface to automatically extract and correlate features. Similarly, queue sorting is used for scheduling thermal power plants participating in load shifting and peak shaving, and stack-pointer-price data structures are also used, reflecting their application in real-time scheduling methods based on priority and price.

[0154] 6. The concept of operators as functions: In data structures, functions reflect electricity prices, active and reactive power, inertia and frequency, carbon emissions, security and stability, and network combinatorial decoupling. These are represented by operators, which are parallel computational function units within an intelligent agent that implement specific functions within a power system, such as electricity prices, system inertia, and frequency. Any function can exist in the form of an operator. For example, a fitted curve function corresponds to a function derived from mapping and summarizing data vectors, and then the logical functional relationships are reinforced through the intelligent agent interface.

[0155] After extracting data features, the data is encapsulated using the aforementioned tools to form a new logical core, known as an operator. Multiple operators form an intelligent agent, and multiple intelligent agents constitute a digital power system. In this operator, the rules governing the flow of information based on logical relationships driven by data features applied to the physical and equipment mechanisms (plant and grid control and equipment) within the power system are detailed in the flowcharts, signaling diagrams, etc., involved in the various embodiments below.

[0156] Based on the three operators of power generation control, clearing, and internal control logic, a correlation is established. Inspired by the feature extraction paper, high-dimensional data vector comparison is used to compare data between the power grid and power plant sides, finding the optimal feature with high correlation weight. This fully reflects the underlying logic and algorithm of the correlation in the operators and the "source-grid" matching degree as a security consideration.

[0157] The three operators—power generation control operator, power grid clearing operator, and internal control logic operator—are functionally described from the perspectives of power source-side power generation mechanism, power grid clearing principle and logic, and security considerations. They describe their respective technical solutions, which meet the inventiveness characteristics of an invention. At the same time, the invention creatively proposes the relationship between these three operators, leading to a number of deeper issues about the construction of digital power systems by multiple agents, such as their symbiosis and migration, until more agents emerge and generate emergence.

[0158] The method of constructing a digital power system by multiple agents essentially involves many operations on data and information flow, including feature extraction, comparison, and matching degree research of data. Information in the flow is controlled according to the mechanism to control the flow direction and construct agents to realize many functions of the digital power system. The key is to construct appropriate operators within the agents so that indicators such as "matching degree" can be constructed to measure various physical mechanisms of the power grid and power plants.

[0159] The operators reflect the production factors of the power grid and are attributed to the power grid. These include functional operators such as electricity price operators, including the electricity price function, the power grid price equilibrium point, and internal control operators, as well as the operators corresponding to electricity charges in the final function. Based on the digital power plant structure of different types of generating units (including their power plant price equilibrium points) and combined with the actual power grid clearing function, a power grid price operator is constructed. Simultaneously, a power grid price equilibrium point is established. This equilibrium point, based on the power grid price, represents the operating condition point where power plant costs and the power grid clearing transaction price are balanced. Its significance lies in serving as a key reference for the power grid in setting electricity prices.

[0160] III. Digitalization and Functionalization

[0161] The embodiment of digitalization and data-driven approaches are deeply involved in several embodiments of this application. For example, in embodiment four, the digitalization is not a single digitalization method. There are differences in large and small sample environments. For the digitalization of large samples, deep learning methods must be used to compare the measured data with the samples to form a workflow. Based on the logical flow in the intelligent agent, repeated deduction and reverse action are used to simulate human reasoning.

[0162] Example 4 collected a small sample of data. By studying the distribution of water levels, it discovered the seasonal cycles and patterns of upstream water levels. Only by digitizing this data and identifying these patterns can quarterly regulation and scheduling strategies be implemented to maximize higher-order data such as power generation revenue, water consumption per unit of power generation, and water assets. Only by finding the correlations between data points, identifying the key factors within these correlations, and analyzing the underlying logic within the market and environment can the most critical data be identified in a data-driven approach. This ensures the correlation and criticality of each item within the selected data vector. Furthermore, other similar low-to-mid-level water level data can be found, and this abstraction of data characteristics can be used to create a data-driven methodology.

[0163] For example, in Embodiment 2, this scheme considers the impact of various power source types on grid power and frequency. A preliminary power allocation curve is derived from the ratio K of the higher-order power difference to the frequency difference. This curve originates from the initial power source's quoted price, power volume, and physical characteristics. The resulting power curve then needs to be fitted using a power source like photovoltaics. This data fitting involves a large amount of real-time data. Based on the degree of fit, the next step is selected under data-driven conditions. If the fit is poor, the output power of other units needs adjustment. Even for grid safety, the K value of the entire grid may be adjusted, allowing the K value or the power output of each power source to be dynamically adjusted and find a balance point under stable frequency regulation. These are all interactions within the intelligent system. At the underlying level, there is data association and abstracted data-driven processing. Parallel computing methods enable this collaboration to respond quickly, achieving a flexibility unattainable by human judgment. Abstracting data patterns replaces the back-and-forth process of abstracting mathematical formulas and approximating function results, simplifying the process of discovering patterns. Digitally transmitting patterns and information directly to subsequent calculations ensures the accuracy and speed of information.

[0164] Figure 1C is a structural diagram of data in the construction of a digital power system provided in this application embodiment, focusing on the data aspect. As shown in Figure 1C, from data elements, data vectors, data functionization to electricity prices, and finally to scheduling algorithms, it directly affects the digitalization of the entire industry chain. The connotation of a digital power system revolves around data. Whether it starts with the data vectors of digitized production factors, to the normalization of production factors, or to the electricity prices and scheduling strategies after calculation, it all reflects the underlying logic of data.

[0165] Safety data vectorization is discussed in Example 4, hydropower cascade scheduling; protection setting digitization is discussed in Example 6, grid structure change; production factor digitization is discussed below, reflecting the digitization and collaboration of multiple production factors; other data vectorization is discussed in Example 3; scheduling algorithms are discussed in Example 2, clearing function, Example 4, hydropower scheduling, and Example 5; artificial intelligence processing of complex scheduling is discussed in Examples 6, 4, and 5; production factor normalization is discussed in Example 2; production and settlement digital functionization is discussed in Examples 3, 2, and 4; electricity price and scheduling functionization is discussed in Examples 2 and 3; and full industry chain digitization is discussed in Example 1.

[0166] V. Reasoning ability and information classification under underlying logic

[0167] The grid-side production factors emphasize the inherent synergy, while the control of specific power plant-side production factors incorporates the digital logic of higher-order control production factors such as hydropower (inertial control) and photovoltaics (active and reactive power control).

[0168] Example 1 describes the production factors of power plants and the power grid, emphasizing their inherent synergy. Essentially, it establishes control over computing power within different functions, such as active and reactive power function computing power groups, power grid inertia function operator groups, power grid load and electricity price function computing power groups, power grid security and stability computing power groups, and power grid carbon footprint computing power groups. These computing power clusters have direct high-speed data connections and can also be nested and invoked, enabling decentralized control through blockchain. These computing power groups are all driven by data from previous production or control of production factors. This constitutes the datafication of operator functions within the intelligent body, which is the core of digital electricity.

[0169] Using computing power clusters as a tool for digitalizing electricity means deeply considering and grasping digital electricity. Professionally, by calculating and measuring the computing power of electricity and the digital world, we can establish a thinking ability similar to 0PENAIO1. This allows intelligent agents to establish a collaborative relationship of function mapping on the grid and power plant sides, and establish related data vector relationships on the data level. Only then can we extensively associate vectors and functions, groups (production factors and characteristics), and intelligent agents (electricity price, active and reactive power, inertia, safety, and carbon footprint of electricity).

[0170] Figure 1D is an implementation architecture diagram of a digital power association provided in an embodiment of this application. The digital power system of the multifunctional intelligent agent in this solution is also built based on the concept of this implementation architecture diagram.

[0171] The following description, with reference to the accompanying drawings and embodiments, details the implementation of the method and related devices for constructing a digital power system based on a multifunctional intelligent agent in this application. Figure 1E is a flowchart of a method for constructing a digital power system based on a multifunctional intelligent agent provided in an embodiment of this application. As shown in Figure 1E, the method includes:

[0172] S101. Extract characteristic production factor data corresponding to the characteristic production factor group from the power data on the power plant side; and extract control production factor data corresponding to the control production factor group from the power data on the power grid side.

[0173] Currently, power data analysis in power systems is generally conducted in isolation. For example, when analyzing a certain indicator, only a very small amount of data directly related to that indicator is considered, thus ignoring the influence of other factors. This leads to insufficient accuracy in the power analysis results, or the content is too simplistic, resulting in low value. In this situation, it is easy to cause ineffective waste of power resources. To construct a digital power system based on multiple agents, this application proposes to coordinate the data characteristics of the power plant side and the data characteristics of the grid side. Specifically, it coordinates the characteristic production factor data of the power plant side and the control production factor data of the grid side, and based on this, constructs a functional operator containing the mapping function of the relationship between the two, and builds an agent accordingly, as detailed in S102-S103. This concept solves the problems of incomplete analysis and overly simplistic data consideration in existing systems. It can more accurately and sensitively capture the relationships between data in the power system and form data assets to meet diverse power function requirements. Thus, it can assist relevant personnel in more efficiently scheduling and controlling the power system, maximizing the effectiveness of the digital power system.

[0174] Figure 1F is a schematic diagram of a power supply scenario provided by an embodiment of this application. As shown in Figure 1F, in practical applications, various types of power plants may communicate with the grid side and transmit power. These different types of power plants can be collectively referred to as the power plant side. In this embodiment, the power plant side can report its own data to the grid side and provide timely feedback; the grid side can perform scheduling on the power plant side, such as peak shaving and frequency regulation. Due to the influence of various factors such as power generation cost, generator performance, uncertainty of power generation by different types of power plants, and changes in load demand, the power plant side often needs to reach a "consensus" with the grid side and then execute corresponding actions according to the grid side's instructions. Taking a thermal power plant as an example, the thermal power plant can further increase or release the heat stored in the thermal power plant according to the grid side's instructions; or the thermal power plant can adjust the following mode of the generator-furnace according to the load requirements. Taking a wind power plant as an example, the wind turbine can increase or decrease the load according to the grid side's instructions and adjust the wind turbine's chamfer, etc. Taking a photovoltaic power plant as an example, the angle of the thyristor can be controlled according to the grid side's load scheduling instructions, thereby changing the power generation. Taking hydropower plants as an example, water volume and water usage costs can be calculated based on the grid-side dispatch curve, thereby controlling the power generation of hydropower turbines. As shown in Figure 1F, in actual power production, close scheduling between the grid and power plants is necessary to achieve better and more stable power supply and maintain cost control for both sides. Based on these practical needs, this application proposes a digital power system that collaboratively constructs a multifunctional intelligent entity from multiple production factors. This system has broad application prospects and the potential for long-term application in power systems. Through the collaborative and digital analysis of multiple production factors, it can improve the accuracy and reliability of current power analysis in power systems. By collaborating multiple production factors, it breaks down data barriers between power plants and the grid, and uses artificial intelligence technology to capture the connections between data from multiple production factors, achieving digital power analysis in the power system field in a more intelligent and automated manner.

[0175] For ease of understanding, the production factor allocation architecture in this application's technical solution is described below with reference to Figure 1G. Figure 1G is a schematic diagram of a production factor allocation architecture provided by an embodiment of this application. In this embodiment, production factors are allocated separately for the power plant side and the power grid side. The basis for allocating production factors on the power plant side is called the characteristic production factor group; the basis for allocating production factors on the power grid side is called the control production factor group.

[0176] In this embodiment, the characteristic production factor group includes multiple levels of characteristic production factors. Specifically, these characteristic production factors can be subdivided into: primary characteristic production factors, secondary characteristic production factors, and tertiary characteristic production factors. Primary characteristic production factors are natural characteristic production factors (or natural indicators), which are production factors that directly reflect natural characteristics. Secondary characteristic production factors are single-system characteristic production factors (or single-system indicators), which are production factors involving individual systems within the power system. Tertiary characteristic production factors are plant system characteristic production factors (or plant system indicators), which are production factors involving the overall system within the power system.

[0177] As shown in Figure 1F, power plants vary in type, including thermal power plants, photovoltaic power plants, wind power plants, and hydropower plants, each employing different power generation methods, utilizing different natural resources, and having different internal system structures. To achieve more accurate digital power analysis, the technical solution in this application can further subdivide each type of power generation into primary, secondary, and tertiary characteristic production factors. Therefore, for each type of power plant, data on the corresponding level of characteristic production factors can be extracted subsequently.

[0178] The following examples illustrate several primary characteristic production factors for different types of power generation. Taking photovoltaic power generation as an example, primary characteristic production factors may include: photovoltaic irradiance. Taking hydropower generation as an example, primary characteristic production factors may include: precipitation (water inflow), etc. Taking thermal power generation as an example, primary characteristic production factors may include: lower heating value of coal, coal consumption, heat consumption, etc. Taking wind power generation as an example, primary characteristic production factors may include: wind speed, etc.

[0179] For thermal power plants, a single system can be a boiler system, an electrical system, etc. For hydropower plants, a single system can be a turbine system, a speed control system, a butterfly valve layer system, etc. For photovoltaic power plants, a single system can be a photovoltaic panel area, a substation area, etc. In addition, these different power plants also share a transmission and transformation system. Correspondingly, secondary characteristic production factors can be boiler efficiency, turbine efficiency, etc., reflecting the production factors of a single system. Tertiary characteristic production factors can be photovoltaic efficiency, hydropower efficiency, thermal power unit efficiency, power plant thermal efficiency (specific heat index), etc., reflecting the overall situation of multiple systems. As can be seen from Figure 1G, the data feature mining level progresses progressively from primary characteristic production factors to secondary characteristic production factors to tertiary characteristic production factors, deepening with each level. Therefore, the feature production factor classification architecture in this embodiment, based on the feature production factor group, can collaboratively extract data on natural features, single-system features, and plant system features, achieving digital power analysis from multiple levels of features.

[0180] In this embodiment, the control production factor group includes multiple aspects of control production factors. Control production factors are production factors that reflect the performance of the power grid and are composed of some key physical quantities in the power system. Control production factors help provide effective theoretical basis for power system control. In this embodiment, for example, the control production factors may include, but are not limited to, the following six aspects: (1) electricity price; (2) inertia and frequency; (3) active and reactive power; (4) carbon emissions; (5) safety and stability; and (6) grid connection and disconnection. It should be noted that the six aspects of control production factors mentioned above are only defined from a macro perspective. In specific implementation, each aspect of control production factors can be further subdivided or derived into more specific indicators. Here, there is no limitation on the quantity and content of control production factors. Based on the classification architecture of control production factors in the control production factor group in this embodiment, data on multiple aspects such as electricity price, inertia and frequency, active and reactive power, carbon emissions, safety and stability, and grid connection and disconnection can be extracted collaboratively to achieve digital power analysis from multiple aspects of power grid concern.

[0181] As previously mentioned, the technical solution of this application pre-plans a group of characteristic production factors and a group of control production factors. The group of characteristic production factors includes multiple levels of characteristic production factors, such as first-level, second-level, and third-level characteristic production factors for various power generation types. The group of control production factors includes multiple aspects of control production factors, such as electricity price, inertia and frequency, active and reactive power, carbon emissions from electricity, safety and stability, and grid connection and disconnection. Based on the aforementioned group of characteristic and control production factors, this step combines power data from both the power plant and grid sides to extract production factor data.

[0182] In one optional implementation, characteristic production factor data corresponding to the characteristic production factor group is extracted from the power data of the power plant, including: extracting primary characteristic production factor data, secondary characteristic production factor data, and tertiary characteristic production factor data corresponding to thermal power generation from the power data of thermal power plants; extracting primary characteristic production factor data, secondary characteristic production factor data, and tertiary characteristic production factor data corresponding to photovoltaic power generation from the power data of photovoltaic power plants; extracting primary characteristic production factor data, secondary characteristic production factor data, and tertiary characteristic production factor data corresponding to hydropower generation from the power data of hydropower plants; and extracting primary characteristic production factor data, secondary characteristic production factor data, and tertiary characteristic production factor data corresponding to wind power generation from the power data of wind power plants.

[0183] In one alternative implementation, control production factor data corresponding to the control-related production factor group is extracted from the power data on the grid side, and a data vector is formed to express a specific function, such as power system stability. This includes extracting electricity price data, inertia and frequency data, power data, carbon emission data, safety and stability index data, and grid connection and disconnection impact data from the power data on the grid side.

[0184] As an example, electricity price data can include electricity prices at different times and under different load demands.

[0185] Inertia and frequency data can include system inertia and frequency. The relationship between system inertia and frequency is an important aspect of power system stability analysis. System inertia typically refers to the rotational inertia of generator units in a power system, reflecting the system's resistance to frequency changes. The following are some key points outlining the relationship between system inertia and frequency: **Inertia Support:** When large-scale load changes or sudden additions or removals of generator units occur in a power system, system inertia can slow down the rate of change of system frequency. This inertia support helps maintain the frequency stability of the power system. **Frequency Change Rate:** The magnitude of system inertia directly affects the frequency change rate. The larger the inertia, the smaller the rate of change of system frequency for a given power imbalance. A smaller rate of change of frequency helps avoid low-frequency load shedding or other stability control measures triggered by excessively rapid frequency changes.

[0186] Voltage and active / reactive power data can include power at different voltages, for example, in the form of (voltage, active power, reactive power). In a power system, voltage and frequency are not directly related, but they are indirectly related to power (active and reactive power, respectively). Voltage drops when reactive power is insufficient and rises when reactive power is excessive; frequency drops when active power is insufficient and rises when active power is excessive.

[0187] Carbon emissions data for electricity can include a carbon footprint, such as the amount of carbon dioxide emitted per kilowatt-hour of electricity generated.

[0188] Safety and stability metrics can include the mean time between failures (MTBF) of the power grid. MTBF is an important indicator of product reliability, especially for electronic products. It represents the average operating time of a product between two consecutive failures under specified conditions and within a specified time frame.

[0189] Image data of grid connection and disconnection can include the degree of impact of each network on other networks.

[0190] In practical implementation, after extracting the data on control factors of production, it can be further converted into a digital vector representation to facilitate network recognition, computation, and processing. The resulting digital vector representation can contain data on one aspect of control factors of production, or it can contain data on multiple aspects of control factors of production. The dimensions and data sources of the digital vector representation can be set according to requirements and are not limited here.

[0191] S102. Based on the power function requirements of the digital power system, construct a target data vector using the extracted characteristic production factor data, and determine the key control production factor data associated with the power function requirements from the extracted control production factor data.

[0192] In this embodiment, an intelligent agent is constructed based on power function requirements and the characteristic production factor data and control production factor data extracted in the previous step. It is understood that multi-level and multi-faceted production factor data were extracted in the previous step. However, some of this data may be highly correlated with power function requirements, which helps in constructing an intelligent agent capable of meeting specific power function requirements. However, some of the extracted data may have a very weak correlation with power function requirements, such as power function requirements being related to electricity prices. In this case, the carbon emission data related to electricity is currently of little significance for electricity price calculation and analysis. Therefore, to avoid the negative and adverse effects of redundant and massive amounts of data on the construction of the intelligent agent, this application extracts and refines the obtained massive amounts of data to facilitate accurate capture of the relationships between data. As reinforcement learning data feature correlations increase and carbon emission data gradually becomes more data-driven, convolutional neural network algorithms will automatically extract these data features and include them in the electricity price operator. This also reflects the mechanism of future inference and emergence capabilities.

[0193] In this embodiment, based on the extracted characteristic production factor data, and considering the specific demand reflected in the power function requirements and the relationship between that specific demand and the characteristic production factors, one or more characteristic production factor data with high relevance to the specific demand are further extracted from the extracted characteristic production factor data (as target characteristic production factors), thereby constructing a data vector. For ease of distinction, this data vector is referred to as the target data vector. When constructing the target data vector, certain construction rules can be used. For example, the construction rules can set the dimensions of the vector and the level or name of the characteristic production factor corresponding to each dimension.

[0194] Table 1 presents some characteristic production factor data extracted from the power generation data of thermal power plants. In Table 1, SHRW represents the weighted specific heat index, in kJ / (kW·h). Ea represents the power supplied (or generated), in kW·h. EP represents the electricity price, in US dollars. SHRcc is the specific heat index, in kJ / (kW·h). U is the coal price, in USD / t. For example, U = $69 / ton. E represents the unit's heat consumption rate, i.e., the heat consumption per kilowatt-hour of electricity, in kcal / kWh. N represents the specific heat index, specifically the heat required per kilowatt-hour of electricity, in kcal / kWh. AUX represents the plant power consumption rate. ECRm is the electricity price per kilowatt-hour, in USD / (kW·h). ECRm represents the electricity charge ratio, i.e., the cost of coal for producing one kilowatt-hour of electricity, in USD / (kW·h).

[0195] Table 1

[0196] If the specific demand reflected in the electricity functional demand is to calculate the electricity price, then multiple target data vectors can be extracted based on the characteristic production factor data shown in Table 1. For example, each row of data in Table 1 can be extracted to construct a target data vector with the structure (operating condition, SHRW, ECRm, Ea, CERm, EP, actual coal consumption per kWh, actual coal cost per kWh, overall coal cost). For example, using the data in the second row of Table 1, target data vector 1 is constructed as (100%, 2138.4, 0.0293, 9267053600 0.0311, 287772056.8, 450.0942, 0.0311, 287802236). Using the data in the third row of Table 1, target data vector 2 is constructed as (95%, 2152.69, 0.0295, 8803700920, 0.0311, 273383454, 453.5357, 0.0313, 275502693). And so on.

[0197] Furthermore, in practical implementation, a correspondence between control production factors and power function requirements can be pre-established. Then, based on specific power function requirements, the corresponding control production factors can be identified as key control production factors in a timely and efficient manner. Based on this, key control production factor data associated with power function requirements can be determined from the massive amount of extracted control production factor data. Similarly, the extracted key control production factor data can also be converted into a digital vector representation.

[0198] S103. Based on the target data vector and the key control production factor data, an artificial intelligence algorithm is used to learn the mapping function between the target data vector and the key control production factor data, a functional operator with the mapping function as its core is constructed, and an intelligent agent is constructed based on the functional operator.

[0199] Figure 1H is a schematic diagram illustrating the relationship between an intelligent agent, operator, function, and production factor data provided in an embodiment of this application. The implementation process of this step will be explained with reference to Figure 1H. As shown in Figure 1H, in this embodiment, a target data vector is obtained based on characteristic production factor data, and key control production factor data is determined based on control production factor data. Next, an artificial intelligence algorithm is needed to learn from the target data vector and key control production factor data, and to infer and analyze the potential data relationships between the characteristic production factor data and key control production factor data behind the target data vector. Through the application of artificial intelligence algorithms, the relationship between the two can be expressed through a mapping function.

[0200] This application proposes constructing functional operators using mapping functions as the core. In the example of Figure 1H, taking agent 1 as an example, it includes multiple functional operators: operator 1, operator 2, and operator 3. Each functional operator loads a mapping function. It is understood that the mapping functions loaded in different functional operators are different because the power function requirements driving the construction of each functional operator may be different, and consequently, the characteristic production factor data and control production factor data used may also be different. By executing this step, the constructed agent corresponds to the aforementioned power function requirements. For example, based on power function requirement 1, agent 1 is finally constructed; based on power function requirement 2, agent 2 is finally constructed.

[0201] In the embodiments of this application, the artificial intelligence algorithms used may be diverse. For example, currently available open-source artificial intelligence algorithms can all be based on the above-described technical concepts, using target data vectors and key control production factor data to infer and analyze the intrinsic relationship between them, thereby obtaining an accurate mapping function. The specific type of artificial intelligence algorithm used is not limited here.

[0202] As shown in Figure 1H, in this embodiment of the application, an intelligent agent can contain one or more functional operators. Taking intelligent agent 1 as an example, it contains three functional operators. Functional operators within the same intelligent agent can be independent of each other. Furthermore, functional operators within the same intelligent agent can also be technically or logically related.

[0203] Figure 1H also shows other intelligent agents besides agent 1, such as agent 2 and agent N shown in Figure 1H. Once these intelligent agents are successively constructed, they form an automated device that integrates the intelligence of multiple agents; therefore, this device can also be considered a "robot." It possesses rich knowledge related to power systems and, by utilizing the functional operators within each intelligent agent, can solve a wide variety of power system analysis problems. With the increasing international cooperation in the power industry, the method of constructing intelligent agents and the architecture of the "robot" proposed in this application represent a breakthrough technology for addressing international power cooperation.

[0204] S104. Add various intelligent agents corresponding to different power function requirements to the digital power system, and use each intelligent agent as a power function implementation unit in the digital power system.

[0205] In practical applications, power function requirements can be expressed through power analysis requests. These requests can be initiated by the power plant, the power grid, or by a third party other than the power plant or the power grid. As long as the entity or device initiating the power analysis request has the authority or qualification to activate a power analysis agent for data analysis, the execution device (e.g., a server or terminal) of the proposed method for creating a digital power system based on multifunctional agents can respond to the request and analyze the matching of the power function requirements carried in the request with one or more existing agents. If no matching agent is found, the steps in the above method need to be called to create an agent or a corresponding operator for the power function requirement.

[0206] Understandably, if the analysis determines that the power function requirements carried in the power analysis request match a pre-built intelligent agent, then that agent is invoked and used as the power function implementation unit. The agent's internal functional operators then respond to the power analysis request. For example, if the demand information carried in the power analysis request indicates the calculation of electricity prices, then the electricity price operator (whose function is to calculate electricity prices) within the power analysis intelligent agent is invoked, and the electricity price is calculated based on the data currently provided to that operator for relevant calculations.

[0207] In this technical solution, the construction of the intelligent agent relies on relevant production factor data from both the power plant and grid sides. Furthermore, the characteristic production factor group includes multiple levels of characteristic production factors, and the control production factor group includes multiple aspects of control production factors. Therefore, it effectively achieves the coordination of multiple levels of characteristic production factors and multiple aspects of control production factors. The intelligent agent constructed based on this effectively utilizes the diverse information in power data to realize the coordinated power functions of multiple production factors, such as power functions for power trading between the grid and power plants, power functions for grid-side dispatch control of power plants, and power functions for grid self-regulation. The intelligent agent can more accurately capture the connections between data, thereby obtaining more accurate and transferable data analysis or calculation results. It can also be understood that the digital power system constructed through this solution forms high-value data assets within the power system, enhancing the analytical value of power data. The constructed digital power system can effectively contribute to domestic power market reform and global power trading.

[0208] In practical applications, it's possible that a single operator cannot satisfy a relatively large analysis objective. In such cases, a possible solution is to invoke multiple operators to satisfy a single analysis objective. This type of situation is described below.

[0209] To address the above issues, in the initial stage of building the intelligent agent, the power analysis demand information can be parsed to obtain multiple sub-demand information and the logical dependencies or digital transformation relationships between them. By parsing and decomposing the power analysis demand information, functional operators can be constructed specifically based on the sub-demand information. Assume that by parsing the power analysis demand information, multiple sub-demand information is obtained, including a first sub-demand information and a second sub-demand information.

[0210] Figure 1I is a flowchart illustrating the implementation of a power analysis intelligent agent according to an embodiment of this application. As shown in Figure 1I, in a specific implementation, the method for constructing a digital power system based on a multifunctional intelligent agent provided in this application embodiment further includes:

[0211] A first target data vector is constructed based on the first sub-demand information, and the first key control production factor data associated with the first sub-demand information is determined; and a second target data vector is constructed based on the second sub-demand information, and the second key control production factor data associated with the second sub-demand information is determined.

[0212] For the specific example shown in Figure 1I, based on the target data vector and the key control production factor data, an artificial intelligence algorithm is used to learn the mapping function between the target data vector and the key control production factor data. A functional operator with the mapping function as its core is constructed, and a power analysis intelligent agent is built based on multiple functional operators. Specifically, this may include:

[0213] Based on the first target data vector and the first key control production factor data, an artificial intelligence algorithm is used to learn the first operator; and based on the second target data vector and the second key control production factor data, an artificial intelligence algorithm is used to learn the second operator; based on logical dependencies or numerical transformation relationships, a transfer function relationship between the first operator and the second operator is constructed; based on the transfer function relationship, the first operator, and the second operator, a power analysis intelligent agent is constructed.

[0214] Thus, a multi-operator power analysis agent comprising a first operator and a second operator has been constructed. Based on this, when a power analysis request is received, and the demand information carried within matches the first and second sub-demand information, the power analysis agent can be invoked based on the transfer function relationship to trigger the first and second operators to generate the final analysis response to the power analysis request. This achieves the collaborative work of multiple functional operators within the power analysis agent. By constructing multiple functional operators with logical dependencies or numerical transformation relationships, and building the power analysis agent accordingly, the agent can handle more complex and diverse power data analysis tasks.

[0215] As mentioned earlier, the transfer function relationship between the first operator and the second operator is constructed based on logical dependencies or numerical transformation relationships. Therefore, in an intelligent agent, different operators also have corresponding logical dependencies or numerical transformation relationships. Below are some dependencies existing in power systems: reactive power and voltage, heat consumption and coal consumption, system inertia and frequency, electricity price and coal price, etc.

[0216] For example, the numerical conversion relationship between different operators can refer to operators that have essentially the same function but differ in terms of transaction or pricing. To meet the interoperability requirements of transaction or pricing, it is necessary to invoke operators to perform numerical conversions at the transaction or pricing level. As an example, operator A calculates the electricity price of country A's power grid, and operator B calculates the electricity price of country B's power grid. When it is necessary to convert the electricity prices of countries A and B, it is necessary to operate according to the numerical conversion relationship.

[0217] In this embodiment, each intelligent agent is configured with multiple types of interfaces to facilitate communication or data exchange with other entities. For example, the intelligent agent is configured with a power plant-side interface, a power grid-side interface, and an intelligent agent interface. The power plant-side interface is used for the intelligent agent to interface with a power plant; the power grid-side interface is used for the intelligent agent to interface with the power grid; and the intelligent agent interface is used for the intelligent agent to interface with other intelligent agents.

[0218] In one possible implementation, the agent can be configured with multiple power plant-side interfaces, each used to interface with different types of power plants or different power plants of the same type. This application does not limit the number of power plant-side interfaces configured for an agent.

[0219] In one possible implementation, an agent can be configured with multiple agent interfaces, each used to interface with different other agents. This application does not limit the number of agent interfaces configured for an agent.

[0220] By invoking the intelligent agent, an analysis response result can be generated in response to the power analysis request. Based on the above description of the interfaces configured on the intelligent agent, in the technical solution of this application, the intelligent agent can send the analysis response result to the power plant through the power plant-side interface; the intelligent agent can send the analysis response result to the power grid through the power grid-side interface; and the intelligent agent can send the analysis response result to other intelligent agents through the intelligent agent interface. It is evident that through the configuration of these various types of interfaces, the intelligent agent has a way to communicate with various external objects and transmit analysis results.

[0221] Furthermore, in this embodiment, a scenario may arise where, after obtaining the analysis response result, the intelligent agent needs to send control commands to the outside world. By configuring various types of interfaces, the intelligent agent can transmit control commands. In one example implementation, the intelligent agent generates a first control command for the power plant, a second control command for the power grid, or a third control command for other intelligent agents based on the analysis response result. It should be noted that the first, second, and third control commands are generated according to specific control requirements. The intelligent agent can send the first control command to the power plant through the power plant-side interface; or send the second control command to the power grid through the power grid-side interface; or send the third control command to other intelligent agents through the intelligent agent interface.

[0222] Figure 1J is a schematic diagram illustrating the relationship between intelligent agents, operators, functions, and production factor data provided in an embodiment of this application. On the left side of Figure 1J, objects of the same level as the concepts of data, functions, operators, and intelligent agents in the technical solution of this application are shown from a biological perspective. On the right side of Figure 1J, the concepts of data, functions, operators, and intelligent agents mentioned in the proposed digital power analysis method for collaborative multi-production factors are shown. The left and right sides of Figure 1J are connected by arrows labeled "symbiotic," reflecting the similarity in the roles or functions of the concepts or objects on both sides.

[0223] As shown on the left side of Figure 1J, protein construction depends on mitochondria, cell construction depends on proteins, and the operation of physiological functions such as human joints (or organs) depends on cells. A complete human body cannot function without diverse and multifunctional joints and organs. It is understandable that mitochondria provide crucial biological characteristics, such as the genetic material deoxyribonucleic acid (DNA) within them. DNA carries the genetic information necessary for the synthesis of RNA and proteins and is an essential biomolecule for the development and normal functioning of an organism. The performance of human joints (or organs) is also inseparable from the genetic information carried by DNA. The data, functions, operators, intelligent agents, and robots shown on the right side of Figure 1J also exhibit a similar progressive relationship. The formation of mapping functions is inseparable from the extraction of characteristic production factor data and the extraction of control production factor data. Mapping functions rely on this data; operators are based on functions; intelligent agents contain one or more operators to achieve the analysis function of power system data; and the resulting robot, possessing diverse intelligent agents, has the ability to handle complex and diverse power analysis needs. All of this is based on the production factor division architecture shown in Figure 1G.

[0224] The foregoing section provided a detailed description of the construction process for a digital power system based on multifunctional intelligent agents. In this application's technical solution, the constructed digital power system incorporates various intelligent agents corresponding to different power function requirements. To facilitate understanding of the constructed digital power system's ability to respond to diverse power function requirements, this application further illustrates the use of operators to implement power functions in conjunction with the following six embodiments.

[0225] In the technical solution of this application, the power function requirement can be: a power function requirement for power trading between the grid side and the power plant side, a power function requirement for realizing the grid side's dispatch control over the power plant side, or a power function requirement for the grid's own regulation. In the technical solution described below in this application:

[0226] In Example 1, the specific power function requirement for power trading between the grid side and the power plant side is to calculate the electricity price. The functional operator is the electricity price operator, which is used to calculate the electricity price. Example 1 introduces the process of using an intelligent agent built based on the electricity price operator in a digital power system. By using an intelligent agent built from the electricity price operator, an electricity price calculation method is implemented, as explained in conjunction with Example 1 below and Figures 2A to 2C.

[0227] In Example 2, the specific power function requirement for power trading between the grid side and the power plant side is power clearing, and the functional operator is the clearing operator, which is used for power clearing. Example 2 describes the process of using an intelligent agent built based on the clearing operator in a digital power system. The difference between the electricity price operator and the clearing operator is that the clearing operator includes a part for calculating the electricity price, but after calculating the electricity price, it implements the clearing curve function according to priority or stack priority, making the clearing operator function more complex. By using an intelligent agent built from the clearing operator, a grid clearing method is implemented, as illustrated in Example 2 below, Figures 3A to 3I and Figures 4A to 4H.

[0228] In Example 3, the power function requirement for grid-side dispatch control over power plant side specifically refers to peak-shaving dispatch in a high-frequency load change scenario for thermal power. The functional operator is a thermal power peak-shaving dispatch operator, which is used to implement peak-shaving dispatch in a high-frequency load change scenario for thermal power. Example 3 describes the process of applying an intelligent agent built based on the aforementioned thermal power peak-shaving dispatch operator in a digital power system. By using an intelligent agent built from the thermal power peak-shaving dispatch operator, a peak-shaving dispatch method for a high-frequency load change scenario for thermal power is implemented, as explained in conjunction with Example 3 below and Figures 5A to 5F.

[0229] In Example 4, the power function requirement for realizing the grid-side dispatch control of the power plant side specifically refers to the water resource dispatch of cascade hydropower stations. The functional operator is the water resource dispatch operator, which is used to realize the water resource dispatch of cascade hydropower stations. Example 4 introduces the process of using an intelligent agent constructed based on the aforementioned water resource dispatch operator in a digital power system. By using an intelligent agent constructed from the water resource dispatch operator, a water resource dispatch method is implemented, as explained in conjunction with Example 4 below and Figures 6A to 6J.

[0230] In Example 5, the power function requirement for realizing grid-side dispatch control over power plants specifically refers to grid dispatch of power plants in a large-scale power system with small-scale grids. The functional operator is the large-scale grid dispatch operator, which is used to realize grid dispatch of power plants in the large-scale power system. Example 5 describes the process of using an intelligent agent built based on the large-scale grid dispatch operator in a digital power system. By using an intelligent agent built from the large-scale grid dispatch operator, a grid dispatch method in a large-scale power system is implemented, as explained in conjunction with Example 5 below and Figures 7A to 7F.

[0231] In Example 6, the power function requirement for grid self-regulation specifically involves adjusting the grid structure. The functional operator is the grid structure adjustment operator, which is used to adjust the grid structure. Example 6 describes the process of using an intelligent agent constructed based on the grid structure adjustment operator in a digital power system. By using an intelligent agent constructed from the grid structure adjustment operator, a method for adjusting the grid structure is implemented, as explained below in conjunction with Example 6 and Figures 8A to 8D.

[0232] Example 1

[0233] In Example 1, the specific power function requirement for power trading between the grid side and the power plant side is to calculate the electricity price. The functional operator is the electricity price operator, which is used to calculate the electricity price. Example 1 describes the process of using an intelligent agent built based on the aforementioned electricity price operator in a digital power system.

[0234] In practice, power plants need to calculate electricity prices and then provide those prices to the power plants themselves. If the power plants and the grid reach an agreement on the electricity price, electricity trading can then take place.

[0235] The following sections, using Figures 2A and 2B respectively, introduce the grid-side and power plant-side processes based on the electricity price operator from the perspectives of the grid side and the power plant side.

[0236] Figure 2A illustrates the grid-side process based on the electricity price operator. First, key control production factor data related to electricity price calculation needs to be extracted from the grid-side power data. Optionally, this key control production factor data can be represented as digital vectors. Next, the grid side also needs to extract target data vectors from the power plant-side power data through feature extraction. It's understood that these data vectors are also related to electricity price calculation. In one example, the performance efficiency related data vector A for unit A is represented as (coal consumption 1, heat rate 1, unit efficiency 1), and the performance efficiency related data vector B for unit B is (coal consumption 2, heat rate 2, unit efficiency 2). These data vectors are aligned, and a comprehensive production factor data vector is obtained (coal consumption 1 + coal consumption 2, heat rate 1 + heat rate 2, unit efficiency 1 + unit efficiency 2). Next, based on the extracted data vectors from the power plant and grid sides, an artificial intelligence algorithm is used to obtain the operator function computing power group. The operator function computing power group contains many functional operators with functions as their core. Having these operator function computing power groups can also be understood as constructing an intelligent agent for power analysis.

[0237] Taking electricity price calculation as an example, the constructed operator function computing power group can be called the power grid load and electricity price function computing power group. In addition, considering other possible power analysis needs, we can also establish power grid inertia frequency computing power groups, active and reactive power function computing power groups, power grid carbon footprint computing power groups, power grid voltage and current function computing power groups, power grid security and stability computing power groups, and power grid connection and disconnection computing power groups, etc. Among them, the active and reactive power function computing power group includes functions used to reflect the active and reactive power performance of the power grid; the power grid inertia function computing power group includes functions used to reflect the performance of power grid stability frequency transformation; the power grid load and electricity price function computing power group includes functions used to calculate the correlation between power grid electricity prices and power plant generation load; the power grid security and stability computing power group includes functions used to calculate power grid step and power flow; and the power grid carbon footprint computing power group includes functions used to calculate the carbon consumption in the production of hydropower, thermal power, photovoltaic, wind power, and other equipment within the power grid.

[0238] Taking the power grid load and electricity price function calculation group as an example, the mapping function, which serves as the core of the electricity price operator, can be expressed as: CERm = U*N / E(1-AUX). CERm is the grid subsidy coefficient for payment per kilowatt-hour. U represents the coal price, which is also the fuel cost subsidy, in US dollars per ton. E represents the unit's heat consumption rate, i.e., the heat consumption per kilowatt-hour of electricity, in kcal / kWh. N represents the specific heat index, in kcal / kWh. AUX represents the plant's power consumption rate. This formula reflects the relationship between the grid subsidy coefficient for payment per kilowatt-hour and characteristic production factors such as coal price, unit heat consumption, specific heat index, and plant efficiency.

[0239] After obtaining the electricity price operator using artificial intelligence algorithms, the electricity price calculation result on the power plant side is obtained when the intelligent agent containing this operator is invoked. It should be noted that since the construction of this operator relies on the power plant's power data, the electricity price calculation result actually reflects a significant portion of the power plant's own cost considerations and profit demands. The grid side will need to further consider this electricity price calculation result in subsequent processing.

[0240] As shown in Figure 2A, after obtaining the electricity price calculation result on the power plant side through the electricity price operator, the calculation result is verified using a financial verification model on the grid side. The main purpose of verifying the electricity price calculation result using the grid-side financial verification model is to ensure the balance and stability of the power plant and the power grid. In one example, the grid-side financial verification model includes: a first verification condition and a second verification condition; the first verification condition is: the actual output of various types of power plants ≥ the grid power volume; the second verification condition is: the sum of the products of the actual output of various types of power plants, the grid electricity price at the corresponding time, and the power generation duration ≤ the product of the grid average price and the grid power volume. If the electricity price calculation result meets the first and second verification conditions, the verification of the electricity price calculation result is determined to be successful; if the electricity price calculation result does not meet either the first or second verification condition, the verification of the electricity price calculation result is determined to be unsuccessful.

[0241] The expression for the first verification condition is: Q 火电 +M 风电 +Q 水电 +M 光伏 ≥N 电量

[0242] The expression for the second verification condition is: Σ(H*Q 火电 *N 各时刻电网电价 )+ΣM 风电 *N 各时刻电网电价 +Σ(H*Q 水电 *N 各时刻电网电价 )+ΣM 光伏 *N 各时刻电网电价 ≤P 电网均价 *N 电量

[0243] In the above expression, Q 火电 M 风电 Q 水电 M 光伏 These represent the actual power output of thermal power plants, wind power plants, hydropower plants, and photovoltaic power plants, respectively. 电量 This represents the amount of electricity generated by the power grid. H indicates the duration of power generation. N 各时刻电网电价 P represents the grid electricity price at the corresponding time. 电网 均价 This represents the average price of the power grid.

[0244] For calculations, thermal power and hydropower are both physical power generation methods with controllable resources, exhibiting determinism; while photovoltaic and wind power are uncontrollable, with strong uncertainty in their natural elements. The different expressions used here, Q and M, represent the difference in data abstraction between the two types of production factors. Thermal and hydropower generation has strong determinism and weak data abstraction, denoted by Q; while wind and solar power generation has strong uncertainty and strong data abstraction, requiring a higher-order data operator function M to represent it.

[0245] If the electricity price calculation result is verified, a suitable and feasible electricity price calculation function for the grid side is further selected from the dispatch strategy library. The dispatch strategy library can be handled by a dedicated operator that provides diverse calculation functions. As mentioned earlier, the construction of the already constructed electricity price operator relies on the power plant's power data; therefore, the electricity price calculation result actually reflects a significant portion of the power plant's own cost considerations and profit requirements. The grid side needs to further consider this electricity price calculation result. The specific consideration method is to utilize the dispatch strategy library. The grid side also needs to select a technically feasible electricity price calculation function from the dispatch strategy library that is suitable for the grid side, and then load this electricity price calculation function into the electricity price operator to update the electricity price operator; the updated electricity price operator is used to obtain the grid side's electricity price calculation result. In this way, the grid side not only references the power plant's electricity price calculation result but also updates the electricity price operator based on its own cost and profit requirements. This process achieves collaborative calculation of the electricity price by both the power plant and the grid side.

[0246] If a grid-side-adaptable and technically feasible electricity price calculation function cannot be selected from the scheduling strategy library, then it is necessary to deeply learn the grid energy consumption average efficiency operator, average electricity price operator, and other important production factor operators, return the learning results as samples, and use artificial intelligence algorithms to learn the electricity price operator.

[0247] In an optional implementation, after loading the electricity price calculation function into the electricity price operator to update the electricity price operator, the collaborative multi-factor digital power analysis method provided in this application embodiment further includes:

[0248] The grid side receives the requested transaction volume and price reported by the power plant to the grid side. It prepares for the transaction between the power plant and the grid side. At this point, the grid side has completed the price calculation function for the power plant. Next, based on the grid side's assessment of grid stability, it determines whether to activate the virtual power plant function. The purpose of activating the virtual power plant function is to access storage capacity, then, according to the principle of optimal cost, determine the cleared output volume of each power generation, and finally, based on artificial intelligence calculations, issue instructions in advance. The grid accesses the storage capacity of the virtual power plant; when there is a power shortage and the price is high, the virtual power plant (battery) is used to discharge; when there is a power surplus and the price is low, the virtual power plant (battery) is charged.

[0249] In the technical solution of this application, the power grid side can record the clearing curves and dispatch curves under different loads or in emergency situations, which can be used as input data for deep learning to continuously verify, learn and imitate the laws and characteristics of the power grid, and serve as a reference for issuing instructions next time.

[0250] If it is determined that the virtual power plant function does not need to be activated, the grid side sends clearing price and power dispatch instructions to each type of power plant based on the grid side's electricity price calculation results. If it is determined that the virtual power plant function needs to be activated, the virtual power plant function is activated to call up storage capacity. Furthermore, the steps for obtaining the electricity price operator using artificial intelligence algorithms are then performed, combining the virtual power plant function's operators with the returned results.

[0251] Figure 2B illustrates the power plant-side process based on an electricity price operator. First, various characteristic production factor data from the power plant are collected to form a data vector. Then, the electricity price operator is trained using key control production factor data from the grid side. The power plant receives the clearing price from the grid side; it then verifies the data using a financial verification model based on the power plant's costs and the clearing price. In an optional implementation, the power plant-side financial verification model includes a third verification condition: the sum of the products of the power plant's actual output, the corresponding grid price, and the power generation duration > the product of the power plant's costs and its actual output. Taking a thermal power plant as an example, the expression for the third verification condition is: Σ(H*Q 火电 *N 各时刻电网电价 )>Σ(H*Q 火电 *P 电厂成本 )

[0252] In the expression for the third verification condition above, Q 火电 This refers to the actual output of the thermal power plant. H represents the power generation duration. N 各时刻电网 电价 Let N be the grid electricity price at the corresponding time. Artificial intelligence is applied to calculate the grid electricity price N at each time point. 各时刻电网电价 As data input, the power plant cost P under different loads calculated by this power plant system is also applied. 电厂成本Substitute the corresponding values ​​into the formula. The expression for this verification condition reflects the cumulative calculation of the actual output of thermal power generation production factors into the electricity bill for grid settlement. This fluctuates with the grid, but for individual power plant cost production factors, it changes with the load factor, ultimately mapping to the power plant cost production factors.

[0253] A single power plant type may have multiple different power sources. As shown in Figure 2B, if the verification result indicates that the verification is passed, it means that the overall verification conditions are met. Further evaluation is needed to determine whether different power sources of the same power plant type can be cleared based on electricity volume and price. If all power sources can be cleared, the cleared electricity volume is executed. Subsequently, the corresponding actions for adjusting the load of the thermal power plant units are taken. If there are power sources that cannot be cleared, it is necessary to calculate the power plant's production costs or adjust the power plant's clearing price, and then re-evaluate whether each power source can be cleared after verification.

[0254] The first, second, and third verification conditions mentioned above are mathematical logic, all of which express some physical constraints in the power system. These verification conditions ensure the stability of the power system.

[0255] Figure 2C shows the signaling diagram between the power plant side, the grid side, and the intelligent agent. The significance of peak shaving in the power system, as shown in Figure 2C, lies in identifying power plants suitable for peak shaving pricing within the grid's stipulated electricity price range. This ensures the grid's revenue and expenditure remain balanced, allowing for simultaneous estimation of the cleared volume and price for various types of power plants within the budget. Clearing is achieved through two buses or algorithms, specifically the technical clearing point of the volume and clearing curve that each power plant can provide under the basic electricity price. The minimum value after the total grid price is allocated to various types of power plants, satisfying the aforementioned inequalities, is the power plant's electricity price and volume clearing point.

[0256] When digital power analysis is required, the measured data is first digitized. Then (based on the power data from the power plant and / or the power grid), the data is cleaned and composed into data vectors to facilitate subsequent parallel computation. The output of the computation is generalized into operators (functions) based on the data model. These operators (functions) perform specific functions. After the agent completes the analysis using these operators, the entire agent logic flow is completed. The established mapping function is used to generate the power analysis results, or the operator output is used to nest and iterate with new agents. Operators are the most important part of the agent, implementing specific functions of the power system, such as frequency operators, active and reactive power operators, and electricity price operators. The electricity price operator is a function. An agent is a process or logic that includes many operators to implement various functions.

[0257] After production factor data is digitized, it undergoes parallel computing. Due to the characteristics of GPUs, data vectors of the same type can be aligned and processed rapidly in parallel. While using neural network algorithms may lose some characteristics of symbolic artificial intelligence algorithms, graph neural networks, with their data tools, are sufficient to preserve and process data features at the input and transmission layers using a self-attention mechanism under probabilistic methods, reflecting data-driven characteristics. To compensate for the loss of symbolic artificial intelligence and to better enable functional computation and control of intelligent agents, the concept of operators is introduced at the output stage. Multiple operators are incorporated into the agent's logic, enabling the agent to acquire corresponding functions. The capabilities of digital power analysis also revolve around the specific capabilities of these operators, making the intelligent agent the basic unit for power system data analysis.

[0258] Example 2

[0259] In Example 2, the specific power function requirement for power trading between the grid side and the power plant side is power clearing, and the functional operator is the clearing operator, which is used for power clearing. Example 2 describes the process of using an intelligent agent built based on the clearing operator in a digital power system.

[0260] In current grid clearing processes, electricity price forecasting for different power plants and power allocation procedures are often involved. However, in practical applications, electricity price forecasting and power allocation are often affected by various factors, leading to poor grid clearing accuracy. To address this issue, this application implements power clearing through clearing modules, which can also be understood as grid clearing. First, low-level data abstraction is performed based on the generation element characteristics of each associated power plant, quantifying the various factors affecting grid clearing into a first data vector, thereby reducing the data complexity of generation element characteristics. Subsequently, long-term trading prices are predicted based on the first data vector and a preset long-term electricity price function library. This predicts the corresponding long-term trading prices of each associated power plant based on its actual generation element characteristics, and establishes a corresponding long-term trading stack. By ensuring the accuracy of long-term trading prices, the accuracy of grid clearing is improved. Furthermore, a spot trading stack is established based on the spot trading prices of each associated power plant, and a first clearing function is determined through the long-term trading stack and the spot trading stack. Accordingly, based on the determined first clearing function, the power supply and demand balance state of the target power grid at the time of clearing under the first clearing function is determined. The first clearing function is then adjusted in real time according to the actual power supply and demand balance state of the target power grid to obtain the second clearing function. In this way, the clearing function of the target power grid can be dynamically adjusted according to its supply and demand state. A specific clearing function is established through a trading stack constructed in real time using long-term and spot trading prices. This allows the grid clearing to effectively consider the impact of power generation factor characteristics on the power generation of associated power plants, as well as the changes in spot trading prices for each associated power plant, thereby determining a more accurate clearing function and improving the accuracy of grid clearing.

[0261] Next, with reference to the accompanying drawings of specific embodiments, a power grid clearing method provided in this application will be described.

[0262] Referring to Figures 3A and 3B, Figure 3A is a flowchart illustrating a grid clearing method provided in an embodiment of this application, and Figure 3B is a signaling interaction diagram illustrating a grid clearing method provided in an embodiment of this application. In Figure 3A, the grid clearing method is implemented on the target grid side, while in Figure 3B, the overall implementation of the scheme is introduced through the interaction signaling between the associated power plant side and the target grid side.

[0263] As shown in Figure 3A, the power grid clearing method specifically includes the following steps:

[0264] S301: Perform low-level data abstraction processing based on the power generation element characteristics of each associated power plant to obtain the first data vector of each power generation element characteristic.

[0265] The power generation characteristics of associated power plants refer to data on various natural and production factors that affect the power generation performance of the plant, such as weather conditions (illuminance, temperature, wind speed, etc.), hydrological water inflow, coal type and coal consumption, etc. In actual power generation scenarios, the power generation performance of associated power plants is often affected by multiple factors such as weather environment, unit equipment, and coal type and coal consumption.

[0266] Therefore, in order to determine the power generation performance of associated power plants as accurately as possible, the grid clearing method provided in this application first needs to perform low-level data abstraction processing on the power generation element characteristics of each associated power plant. This abstracts the power generation element characteristics that have a significant impact on the power plant's power generation performance from the complex power generation element characteristics, and integrates these power generation element characteristics into a first data vector. Simultaneously, the first data vectorization of the power generation element characteristics provides a data foundation for subsequent adjustments to the preset long-term electricity price function library. Through the preset long-term electricity price function library and the first data vector, the long-term trading electricity price of associated power plants can be predicted more accurately in subsequent processes, thereby improving the accuracy of grid clearing.

[0267] Specifically, taking a thermal power plant as an example of an associated power plant, after low-level data abstraction, the first data vector of a thermal power plant can be its corresponding thermal performance data characteristics, such as (unit heat rate, coal consumption, plant power consumption rate). Similarly, when the associated power plant is a photovoltaic power plant, its corresponding first data vector can also be the photoelectric conversion efficiency characteristics, such as (illuminance, power generation, photoelectric conversion efficiency).

[0268] Next, with reference to the accompanying drawings of specific embodiments, the process of performing low-level data abstraction on the characteristics of power generation elements in this step will be described.

[0269] Referring to Figure 3C, this figure is a schematic flowchart of another grid clearing method provided in an embodiment of this application, which specifically includes the following steps:

[0270] S1101: Based on the characteristics of each of the power generation elements, establish an energy consumption data model for each of the associated power plants; the energy consumption data model is used to evaluate the power generation performance of the associated power plants.

[0271] Specifically, in the process of low-level data abstraction based on the power generation element characteristics of each associated power plant, it is necessary to convert the collected power generation element characteristics into primary indicators (natural indicators), and based on the primary indicators, establish energy consumption data models of secondary and tertiary indicators related to the associated power plants. These indicators include parameters such as energy consumption and power generation efficiency. By constructing energy consumption data models for each associated power plant, the power generation performance of the associated power plants under different conditions can be reflected, thereby comprehensively evaluating the impact of various power generation element characteristics on the power generation performance of the associated power plants.

[0272] Meanwhile, by constructing an energy consumption data model based on the power generation element characteristics of each associated power plant, a mapping can be established between the data corresponding to the power generation element characteristics and specific functions in the preset long-term electricity price function library, thus initially establishing the relationship between data and functions.

[0273] S1102: The energy consumption data model is used to perform low-level data abstraction processing on the characteristics of each power generation element to obtain the first data vector of each power generation element characteristic.

[0274] After constructing the energy consumption data model of each associated power plant, the energy consumption data model is used to perform low-level data abstraction processing on the characteristics of each power generation element. In this way, the power generation element characteristics that can accurately characterize the power generation performance of the associated power plants are extracted from the complex power generation element characteristics, and these power generation element characteristics are integrated into the first data vector corresponding to each associated power plant.

[0275] The resulting first data vector can accurately describe the key characteristics of the associated power plant during the power generation process. It will also be used to adjust the preset long-term electricity price function library and applied in long-term electricity price forecasting, thereby achieving dynamic and rapid long-term electricity price forecasting, so as to quickly adjust the clearing function of the target power grid.

[0276] In one possible implementation, after constructing the energy consumption data model of each associated power plant, the weights of each indicator are set in the energy consumption data model to highlight the importance of different power generation elements in power output, thereby effectively improving the accuracy of subsequent function adjustments to the preset electricity price function library.

[0277] The above is the specific process for low-level data abstraction of power generation element characteristics in step S301. Next, we will continue to introduce the power grid clearing method shown in Figure 3A.

[0278] S302: Based on the preset long-term electricity price function library and each of the first data vectors, perform long-term transaction electricity price prediction to establish a long-term transaction stack.

[0279] The pre-defined long-term electricity price function library stores multiple operator functions for calculating the long-term transaction electricity prices of associated power plants. By using the electricity price operator functions for different power plants in the pre-defined long-term electricity price function library, and combining them with a first data vector determined by the characteristics of each power generation element, the index parameters represented by the first data vector can be imported into the corresponding operator functions, thereby calculating the long-term transaction electricity prices of each associated power plant. This allows for the prediction of long-term transaction electricity prices for associated power plants.

[0280] For example, taking a hydropower plant as an example of an associated power plant, its corresponding long-term electricity price operator function is:

[0281] PTOPE (Long-Term Trading Price) = A + B + C + D + E = Am + Bm + Cm + Dm + Em = (TOPEm - PCm) * CCRm + (TOPEm - PCm) * FOMRm + (TOPEm - PCm) * HFCm + NEOm * VOMRm + (TOPEm - PCm) * CCRTm = (TOPEm - PCm) * (CCRm + FOMRm + HFCm + CCRTm) + NEOm * VOMRm

[0282] In the formula, A represents the capital cost recovery factor, B represents the fixed operation and maintenance cost factor, C represents the hydropower facility factor, D represents the variable operation and maintenance cost factor, E represents the special facility factor, and Am, Bm, Cm, Dm, and Em represent the factors (rupees) for each billing period.

[0283] TOPEm represents the amount of electricity (kWh) that is either taken or paid during the billing period. It is obtained through the annual TOPE, which needs to be confirmed by the target grid and the relevant ordering parties.

[0284] CCRm represents the capital cost recovery charge rate (Rp / kWh) applicable to the billing period;

[0285] PCm = Payment Credit for the Billing Period (kWh). Payment credit PCm exists only at the end of each billing period, and only when the net output for the billing period is less than TOPEm.

[0286] Here: PCm = TOPEm - NEOm - DCm - WLCm;

[0287] NEOm represents the net output during the billing period; NEOm during the billing period should not exceed TOPEm.

[0288] DCm represents the electricity credit for the billing period, which should be equal to the electricity generated by the non-power plant (kWh) according to the dispatch instructions of the target grid plus the deemed electricity delivered or generated; if the power supply system of the target grid cannot accept all the electricity generated by the associated power plant, the target grid should issue appropriate dispatch instructions to the power plant.

[0289] FOMRm = Fixed operating and maintenance costs (in rupees) applicable during the billing period, collected at a rate (Rp / kWh), where HFCm represents the allowances for: (i) a portion of the operating and maintenance costs of the hydroelectric facility, (ii) the watershed coordinating contribution, and (iii) the water charge. All three are expressed in Rp / kWh.

[0290] VOMRm represents the variable operating and maintenance cost recovery rate (Rp / kWh) for the billing period, i.e.:

[0291] VOMRm = VOMRFm + VOMRLm;

[0292] In the formula, VOMRFm represents the variable operating and maintenance cost recovery charge rate (Rp / kWh) applicable to the billing period (in non-rupees), and VOMRFm represents the variable operating and maintenance cost recovery charge rate (Rp / kWh) applicable to the billing period (in rupees).

[0293] CCRTm represents the capital cost recovery charge rate (Rp / kWh) applicable to the special facility during the billing period.

[0294] As can be seen from the above formula, by importing the first data vector containing the power generation element characteristics of a hydropower plant into its corresponding long-term electricity price operator function, the long-term transaction electricity price of the hydropower plant within a certain future time period can be predicted. By importing the first data vector of each power plant into the long-term transaction electricity price operator function corresponding to each associated power plant in the preset long-term electricity price function library, the long-term transaction electricity price of each power plant can be predicted.

[0295] It is understandable that in actual long-term electricity price forecasting scenarios, the characteristics of power generation factors of related power plants in different regions and countries are different, and their corresponding electricity price calculation formulas will also be different. This embodiment will not elaborate on this.

[0296] Furthermore, by ranking the long-term transaction prices of each power plant, the stack priority in the long-term transaction stack can be determined, thereby establishing the long-term transaction stack. For example, Figure 3D shows a schematic diagram of a long-term transaction stack, where the stack is ranked based on the predicted long-term transaction prices of each associated power plant.

[0297] It is understandable that in practical application scenarios, the long-term trading volume between the power grid and its associated power plants is determined by the long-term trading contracts signed in advance between the two parties. In the process of constructing the long-term trading stack, the long-term trading volume indicated in the long-term trading contract can also be used as the weight of the stack. The more trading volume and the lower the long-term trading price, the higher the priority of the stack. When the long-term trading price is the same, in order to ensure the clearing stability of the power grid, the long-term trading volume needs to occupy more weight than the long-term trading price. The amount of trading volume is used as the key factor in the construction of the long-term trading stack. This embodiment will not elaborate on this.

[0298] Therefore, by using the first data vector corresponding to each associated power plant and the preset long-term electricity price function library, the specific long-term transaction price of each associated power plant under the influence of the current power generation factor characteristics can be predicted, and a long-term transaction stack can be established through the long-term transaction electricity price of each associated power plant.

[0299] Next, the process of establishing the long-term transaction stack in step S302 will be described with reference to the accompanying drawings of specific embodiments. Referring to Figure 3E, this figure is a schematic flowchart of another grid clearing method provided in this application embodiment, specifically including the following steps:

[0300] S2201: Using the first data vector, the function indicators of the preset long-term electricity price function library are adjusted in a data-driven manner to obtain a dynamic electricity price function library.

[0301] As mentioned earlier, the preset electricity price function library contains pre-stored operator functions for calculating the long-term transaction electricity price of each associated power plant. In practical applications, the long-term transaction electricity price of an associated power plant is determined by its own power generation cost, which varies significantly across different regions, time periods, and even seasons. Therefore, to accurately predict the long-term transaction electricity price of associated power plants, it is necessary to use a first data vector generated from the characteristics of power generation elements to perform data-driven adjustments on the function indicators in the preset electricity price function library. This adjustment aims to refine the function indicators in each operator function, thereby improving the prediction accuracy of the long-term transaction electricity price.

[0302] Here, we take a thermal power plant as an example of a related power plant. Taking the first data vector of the thermal power plant as (unit heat consumption rate, coal consumption, plant power consumption rate, electricity price) as an example, its first data vector is: (Pm, SHRW, GCVs, AUX, CERm), and its corresponding long-term trading electricity price operator function is: CERm=Pm*SHRW / GCVs(1-AUX)=0.031053242USD / KWh(1)

[0303] In the formula, Pm represents the coal price USD, SHRW represents the specific heat index of the unit Kcal / KWh, GCVs represents the higher heating value of coal Kcal / Kg, AUX represents the plant power consumption rate, in %, and CERm represents the electricity price, in USD / KWh. In formula *(1), SHRW is set to 2138.4Kcal / KWh.

[0304] The electricity price operator function is adjusted using the first data vector. The adjusted electricity price operator function is: CERm=Pm*SHRW(1+AGE) / GCVs(1-AUX)=0.031947846USD / KWh(2)

[0305] In the formula, AGE represents the unit loss, in percentage. Other values ​​are the same as above. When AGE is set to 5.4%, SHRW = 2210.14 kcal / kWh.

[0306] Specifically, please refer to Figures 3F and 3G. Figure 3F shows a parameter representing the thermal performance of the unit under full load conditions provided by an embodiment of this application. Figure 3G is a schematic diagram of parameters considering coal price and long-term trading electricity price based on the data shown in Figure 3F. It can be seen that after the specific heat index is changed due to the performance degradation of the unit, the loss needs to be taken into account in the unit electricity price change formula, so that (2) replaces formula (1), and the electricity price changes from 0.031 USD / KWh to 0.032 USD / KWh.

[0307] Therefore, by using the first data vector determined in real time by each associated power plant, the long-term electricity price operator function in the preset long-term electricity price function library for each associated power plant can be adaptively adjusted, thereby ensuring the accuracy of long-term transaction electricity price prediction.

[0308] S2202: Based on the dynamic electricity price function library and the characteristics of each of the power generation elements, long-term transaction electricity price prediction is performed to obtain the long-term transaction predicted electricity price of each of the associated power plants.

[0309] Correspondingly, by dynamically adjusting the preset long-term electricity price function library, the characteristics of each power generation element are used as input data in the adjusted dynamic electricity price function library, and the long-term transaction forecast electricity price of each associated power plant can be calculated.

[0310] S2203: Establish the long-term transaction stack by predicting electricity prices through the long-term transactions of each of the associated power plants.

[0311] Therefore, the first data vector generated by the actual power generation characteristics of each associated power plant can effectively adjust the function operators in the preset electricity price function library. Thus, the adjusted dynamic electricity price function library can be used to predict the actual long-term transaction electricity price for the associated power plants, thereby ensuring the accuracy of the long-term transaction electricity price prediction and obtaining the long-term transaction stack.

[0312] S303: Establish a spot trading stack based on the spot trading electricity price of each of the associated power plants, and determine a first clearing function through the long-term trading stack and the spot trading stack.

[0313] During the establishment of the long-term trading stack, the target power grid synchronously receives spot trading prices from associated power plants. The target power grid can establish corresponding spot trading stacks based on the magnitude of the spot trading prices from each associated power plant. In this way, the first clearing function of the target power grid can be determined through the spot trading stack and the long-term trading stack.

[0314] For the spot trading stack and the long-term trading stack, please refer to Figure 3H. Figure 3H is a schematic diagram of the stack structure of a spot trading stack and a long-term trading stack provided in an embodiment of this application. As shown in the figure, the stack is divided into three levels according to the electricity price, and further divided into top, middle, and bottom levels according to the progressively increasing electricity price. Each level considers different energy types (such as photovoltaic and wind power) and market types (spot trading and long-term trading). The adjustment mechanism for the stack is shown on the side, which balances between spot trading and long-term trading by reducing or increasing the trading volume, thereby optimizing the overall electricity price and power distribution.

[0315] In practical applications, spot market electricity prices for various power plants often take multiple forms, such as hourly quotes and block quotes. The spot market electricity price is calculated using a formula set internally by the power plant. Taking a hydropower plant's hourly quote calculation as an example, the formulas are as follows:

[0316] Hydropower plant: Q m3 =PR 实 ×P m (1.2)

[0317] In the formula, m is the m-th hydropower station in the cascade hydropower station series, and i is the i-th generating unit in the m-th hydropower station; m and i are both integers not less than 1; Q in formula (1.1) m2 This represents the actual revenue corresponding to the potential electrical energy of the hydropower station at the current moment; Q in (1.2) of the objective formula. m3 G represents the actual power generation revenue corresponding to the actual electrical energy generated by the hydropower station at the current moment. i H represents the rated power of the i-th unit. im F represents the duration of the water level of the i-th unit. iJm H represents the derating power of the i-th unit. iZm P represents the duration of the water level interruption during the i-th generating unit's water level period. im The power generation of the i-th unit during the water level period, PR 实 P represents the actual electricity price traded by all generating units. m Electricity generated by all generating units during the water level period.

[0318] As mentioned earlier, the preset electricity price function library contains pre-stored operator functions for calculating the spot trading prices of various associated power plants. In practical applications, the spot trading price of associated power plants is determined by their own generation costs. These generation costs vary significantly across different regions, time periods (due to the degradation of unit performance over time), and even seasons. Therefore, to accurately predict the long-term trading prices of associated power plants, it is necessary to use a first data vector generated from the characteristics of generation elements to perform data-driven adjustments to the function indicators in the preset electricity price function library. This adjustment aims to refine the function indicators in each operator function, thereby improving the accuracy of long-term trading price predictions.

[0319] Furthermore, the establishment of a spot trading stack requires construction based on two dimensions: the spot trading electricity price of the associated power plant and the amount of electricity generated. The following will describe the establishment process of the spot trading stack with reference to specific embodiment figures. See Figure 3I, which is a flowchart illustrating a spot trading stack establishment method provided in this embodiment, specifically including the following steps:

[0320] S3031: Based on the first data vector of each associated power plant, perform power prediction to obtain the predicted power of each associated power plant.

[0321] As mentioned above, the first data vector corresponding to the associated power plant contains multiple power generation factor features that affect the power generation efficiency of the associated power plant. Based on the first data vector, power generation prediction can be performed to predict the power generation of the associated power plant within a certain period of time.

[0322] Specifically, in the process of power generation prediction based on the first data vector, a neural network model can be used to predict the power generation of associated power plants within a certain time period. Taking a wind farm as an example, see Figure 4A, which is a schematic diagram of power generation prediction based on a wind farm according to an embodiment of this application.

[0323] As shown in the figure, global forecast data, regional real-time data, and on-site observation data of wind farms over a historical period are used as training data for the model. This trained neural network model then predicts the power generation of wind farms. Correspondingly, the model can output meteorological forecasts and corresponding power predictions for wind farms. By using meteorological and power prediction data over a certain period, the model can effectively predict the power generation of wind farms based on the main characteristics affecting their power generation performance.

[0324] Furthermore, for power generation forecasting of associated power plants, time series forecasting models (ARIMA) or machine learning methods (such as random forests and XGBoost) can be used. For the first data vector with its complex data structure in this application, deep learning models such as LSTM (Long Short-Term Memory) can be used to decipher the temporal relationships in the sequence data.

[0325] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) architecture designed to address the vanishing and exploding gradient problems faced by ordinary RNNs when processing long sequences of data. By introducing memory units and gating mechanisms, LSTM can effectively capture long-term dependencies in sequences.

[0326] S3032: Establish a preliminary spot trading stack based on the predicted power generation of each of the associated power plants.

[0327] After obtaining the predicted power generation for each associated power plant, the priority of each plant in power dispatch can be determined based on the magnitude of the predicted power generation, thus establishing a preliminary spot trading stack. For example, taking wind power, solar power, hydropower, thermal power, and nuclear power as associated power plants, let the predicted power generation for each plant be (100, 50, 30, 60, 20). Based on the predicted power generation of each associated power plant, a preliminary spot trading stack can be established, namely (wind power, thermal power, solar power, hydropower, nuclear power).

[0328] S3033: Based on the spot trading electricity price of the associated power plant, adjust the preliminary spot trading stack to obtain the spot trading stack.

[0329] Based on the preliminary spot trading stack established according to the predicted power generation of each associated power plant, the spot trading price calculated using the spot price calculation formula within each associated power plant is further adjusted to transform the two-dimensional preliminary spot trading stack into a three-dimensional spot trading stack including spot trading prices. Thus, when an associated power plant generates insufficient or excessive power, the upper-level stack is removed and reconstructed. Starting from the power generation of the associated power plants and the actual spot trading price, the spot trading stack is dynamically adjusted, thereby improving the accuracy of grid clearing.

[0330] For example, a spot trading stack can be seen in the spot trading stack diagram shown in Figure 11.

[0331] It is important to note that in the scenario of establishing a spot trading stack, the weight of the predicted electricity volume in the stack is greater than the weight of the spot trading price. That is, when two related power plants have the same spot trading price, the related power plant with the larger predicted electricity volume is placed in a relatively higher priority position in the spot trading stack to ensure the power supply stability of the target grid.

[0332] S104: Determine the power supply and demand balance state of the target power grid when it is cleared under the first clearing function, based on the first clearing function.

[0333] Accordingly, by controlling the target power grid to perform power clearing based on a determined first clearing function, the power supply and demand balance of the target power grid can be determined according to the total generation and total load of the target power grid at the time of clearing based on the first clearing function. When the total generation of the target power grid is greater than the total load, it indicates that the target power grid is in an over-generation state, which means that the current generation of the target power grid exceeds the actual power demand, and there is a problem of power surplus. Conversely, when the total generation of the target power grid is less than the total load, it indicates that the target power grid is in an under-generation state, which means that the current generation of the target power grid cannot meet the actual power demand, and power supply needs to be supplemented in a timely manner.

[0334] For details, please refer to Figure 12, which shows a schematic diagram of a grid clearing function. In the figure, the horizontal axis represents the change in power supply, and the vertical axis represents the change in electricity price. The power dispatch priority shown in the figure is: wind power / solar power > hydropower with reservoir capacity > nuclear power > coal power > gas power.

[0335] When the electricity supply curve is higher than the electricity demand curve, it indicates that the electricity supplied by the grid exceeds demand, which may lead to a further decrease in electricity prices. Conversely, when the electricity supply curve is lower than the electricity demand curve, electricity demand exceeds supply, which may lead to an increase in electricity prices.

[0336] By determining the power supply and demand balance state of the target power grid under the first clearing function, it is possible to characterize whether the target power grid is in an under-generated or over-generated state under the current first clearing function. This allows for subsequent control of the long-term trading stack and spot trading stack based on the actual power supply and demand balance state of the target power grid, thereby achieving clearing adjustments for the power grid and ensuring the accuracy of the power grid clearing.

[0337] S305: Adjust the first clearing function according to the power supply and demand balance state to obtain a second clearing function, and perform power clearing based on the second clearing function.

[0338] Finally, based on the over-generation or under-generation state of the target power grid under the first clearing function, the first clearing function is adjusted. This allows the clearing function of the target power grid to be adjusted according to the actual power generation situation of the target power grid, thereby improving the accuracy of grid clearing.

[0339] Next, with reference to the accompanying drawings of specific embodiments, the adjustment methods for the first clearing function under different power supply and demand balance states will be described. Referring to Figures 4D and 4E, Figure 4D is a flowchart illustrating another grid clearing method provided in this application embodiment, and Figure 4E is a flowchart illustrating the adjustment method for the first clearing function when the target grid is in an over-generating state, which specifically includes the following steps:

[0340] S3051: When the power supply and demand balance state is the over-generation state, the characteristics of each power generation element are subjected to high-order abstraction processing through a preset deep learning model to obtain the second data vector of each associated power plant.

[0341] In practical applications, long-term electricity trading prices are often determined by long-term contracts between the target power grid and associated power plants, while spot electricity trading prices are often affected by immediate supply and demand conditions. Therefore, long-term electricity trading prices are often lower than spot electricity trading prices. When the target power grid's electricity supply and demand balance is in a state of over-generation, the total power generation of the target power grid exceeds its total load demand.

[0342] To minimize the cost of power dispatch and improve the accuracy of grid clearing, when the target grid is in an over-generating state, since the total generation of the target grid is greater than the total charge, the proportion of long-term trading volume in the clearing function can be increased by increasing the first long-term trading volume in the clearing function. This effectively reduces the cost of power clearing and improves the accuracy of power clearing.

[0343] Therefore, in order to ensure the accuracy of long-term trading volume adjustments, it is necessary to use a pre-defined deep learning model to perform high-level data abstraction on the power generation characteristics of associated power plants, and obtain a second data vector that affects the power generation efficiency of each associated power plant.

[0344] Unlike low-level data abstraction processing that focuses on the characteristics of power generation elements through energy consumption data models, which only involves feature selection and combination of power generation element characteristics, high-level data abstraction processing involves combination calculations between the characteristics of various power generation elements to calculate the data vector that has a core impact on the power generation performance of related power plants.

[0345] For a specific example, please refer to Figure 4F, which illustrates the calculation of a second data vector using a deep learning model. The figure uses a hydropower plant as an example of a related power plant. Parameters such as power generation and water consumption represent the power generation characteristics of the hydropower plant. The plant's Top, Qm', and average plant revenue are parameters calculated using power generation and water consumption. This process of calculating these parameters based on the power generation characteristics is the aforementioned process of performing high-level abstraction processing on the power generation characteristics using a preset reinforcement learning model to obtain the second data vector. In the preset reinforcement learning model provided in this application embodiment, various parameter calculation formulas of different types are pre-set. When performing high-level data abstraction processing on the power generation characteristics using the preset reinforcement learning model, parameters are calculated using the pre-set calculation formulas and the power generation characteristics. The results are then integrated to obtain the second data vector.

[0346] Compared to linear models, deep learning possesses stronger nonlinear modeling capabilities, enabling it to simulate the complex nonlinear relationships between the characteristics of various power generation elements in a connected power plant. Therefore, by pre-setting a deep learning model for high-order data abstraction of power generation element characteristics, complex features can be further extracted from the original data to obtain a second data vector that accurately characterizes the power generation performance of the connected power plant.

[0347] S3052: Determine the long-term transaction volume increase value based on each of the second data vectors and the preset long-term electricity price function library.

[0348] S3053: Adjust the first long-term trading volume according to the increase in the long-term trading volume to obtain the second clearing function.

[0349] Furthermore, after determining the second data vector for each associated power plant, the function indicators in the preset long-term electricity price function library are adjusted based on the second data vector. This can change the specific indicator values ​​of the long-term electricity price function, thereby determining the new long-term trading volume of the target power grid. By comparing this new long-term trading volume with the long-term trading volume in the first clearing function, the increase in the long-term trading volume can be determined, and the corresponding second clearing function can be obtained.

[0350] As shown in Figure 4D, when the target power grid is in an over-generation state, by using a pre-set deep learning model to perform high-order data abstraction processing on the generation element characteristics of the associated power plants, the increase in long-term transaction volume can be determined, and the subsequent long-term transaction stack can be adjusted, thereby achieving the adjustment of the first clearing function and obtaining the second clearing function.

[0351] Next, with reference to the accompanying drawings of specific embodiments, the process of adjusting the first clearing function when the target power grid is in an under-generating state will be described. Referring to Figure 4G, this figure is a schematic flowchart of a method for adjusting the first clearing function when the target power grid is in an under-generating state, provided by an embodiment of this application, specifically including the following steps:

[0352] S3054: When the power supply and demand balance state is the under-generation state, for the first generation curve of the target power grid when cleared based on the first clearing function, the first generation curve is fitted by a preset fitting function to obtain a fitted generation curve.

[0353] When the target grid is in a state of undergeneration, it indicates that the total generation capacity of the target grid is lower than its total load capacity. Understandably, in a dual trading stack consisting of a long-term trading stack and a spot trading stack, the long-term trading price in the long-term trading stack is often slightly lower than the spot trading price in the spot trading stack, thus allowing for priority dispatch of long-term trading power. In this situation, if the target grid remains in a state of undergeneration, it means that even if it fully utilizes the long-term trading power in the long-term trading stack, it still cannot meet its load demand.

[0354] Therefore, under these circumstances, it is necessary to increase the spot trading volume to meet the electricity load demand. To ensure the accuracy of the increase in spot trading volume, before calculating it, it is first necessary to fit the first generation curve of the target power grid when clearing according to the first clearing function, based on a preset fitting function. This transforms the complex and discrete generation data into a continuous function model, thereby simplifying the data storage and analysis process and improving data processing efficiency. Simultaneously, the fitted generation curve obtained after fitting can more clearly identify trends and patterns in the generation data, facilitating subsequent optimization of the clearing function.

[0355] Specifically, the formula for its preset fitting function is as follows: PR*S=P; PR=A+BH T -CH T 2 ;

[0356] In the formula, PR represents the power generation conversion efficiency of the power plant, and H T denoted as the power generation efficiency adjustment factor, S represents the primary index affecting the power generation efficiency of associated power plants, P represents the fitting coefficient, and A, B, and C represent the characteristic constants calculated by the deep learning model, which are used to characterize the performance characteristics of different power plants.

[0357] In the above formula, H T The power of represents the degree of influence of this feature on the power generation performance of the associated power plant. Correspondingly, the formula for H...T The positive or negative sign at the beginning indicates the positive or negative impact of the corresponding feature on the power plant's generation efficiency. Therefore, the above formula can effectively improve the fitting accuracy of the first generation curve, thereby ensuring the accuracy of grid clearing.

[0358] Understandably, the data type corresponding to S will differ depending on the specific power plant involved. For example, when PR represents the power generation conversion efficiency of a photovoltaic power plant, S represents the irradiance of that photovoltaic power plant. Similarly, when PR represents the power generation conversion efficiency of a thermal power plant, S represents the coal quality of that thermal power plant.

[0359] By calculating the fitting coefficient P of all associated power plants corresponding to the target power grid, the first power generation curve of the target power grid can be effectively fitted, thereby improving the accuracy of the first power generation curve.

[0360] Specifically, please refer to Figure 4H, which is a schematic diagram of a target power grid fitted power generation curve provided in an embodiment of this application. As shown in the figure, the topmost power generation curve is a preset power generation curve, and the curve closely matching the preset power generation curve is the fitted power generation curve after the target power grid clears based on the first clearing function and has undergone fitting processing. It can be seen that after fitting the first power generation curve using the preset fitting function, the fitted power generation curve is closer to the preset power generation curve. By comparing the preset power generation curve and the fitted power generation curve, the amount of spot trading that needs to be supplemented can be accurately determined, thereby ensuring the stability of power grid clearing.

[0361] S3055: Based on the fitted power generation curve and the preset planned power generation curve, determine whether the first spot trading volume meets the preset planned power generation curve.

[0362] Subsequently, by comparing the fitted power generation curve with the pre-set planned power generation curve, it can be determined whether the first spot trading volume in the first clearing function can meet the pre-set planned power generation curve. The pre-set planned power generation curve represents the power generation curve expected by the target power grid for the current electricity load. By comparing the two, it can be determined whether the first spot trading volume of the target power grid in the first clearing function is sufficient.

[0363] Referring to Figure 4H, as shown, the fitted power generation curve is lower than the preset power generation curve, meaning the power generation of the fitted curve is lower than the preset curve. In this case, there is a certain gap between the fitted curve and the preset curve. It can be determined that the power generation curve of the target grid under the first clearing function cannot meet the requirements of the preset curve. Therefore, it can be concluded that the first spot trading volume in the first clearing function cannot meet the preset planned power generation curve, requiring additional spot trading volume to supplement it.

[0364] S3056: When the first spot trading volume does not meet the preset planned power generation curve, the preset long-term electricity price function library is modified by a preset reinforcement learning model to determine the reduction value of the long-term trading volume.

[0365] Correspondingly, when the initial spot trading volume does not meet the preset planned generation curve, it indicates that the target grid's initial spot trading volume cannot meet the current electricity load. Therefore, in order to meet the current electricity load by increasing the spot trading volume, it is necessary to modify the preset long-term electricity price function based on the preset reinforcement learning model to obtain the reduction value of the long-term trading volume. In this way, the increase value of the spot trading volume can be determined by reducing the proportion of the long-term trading volume, thereby ensuring the smooth clearing of the grid.

[0366] The pre-defined reinforcement learning model can employ MARL (Multi-agent Reinforcement Learning). MARL is a method for handling reinforcement learning problems in multi-agent environments. Each participant (such as different associated power plants) can be considered an agent, which needs to adjust its bidding strategy based on real-time market data and the behavior of other agents. MARL can help these agents learn and adapt to complex and dynamically changing market conditions to improve the clearing accuracy of the power grid.

[0367] Specifically, during the process of modifying the preset long-term electricity price function library, the first long-term transaction price function with the greatest impact on the overall long-term transaction price can be selected from the preset long-term electricity price function library through the screening mechanism of the reinforcement learning model.

[0368] Subsequently, by adjusting the function index in the first long-term electricity price function and increasing the function index, the long-term electricity price can be increased, thereby reducing the long-term trading volume and obtaining the reduction value of the long-term trading volume.

[0369] Taking a thermal power plant as an example, as mentioned earlier, the long-term trading price calculation function for a thermal power plant is:

[0370] By adjusting the function indicators in the long-term electricity price calculation function, the long-term electricity price can be increased, thereby reducing the long-term trading volume and obtaining the reduction value of the long-term trading volume.

[0371] S3057: Adjust the proportion of the spot trading volume in the first clearing function according to the decrease in the long-term trading volume to obtain the increase in the spot trading volume.

[0372] S3058: Adjust the first spot trading volume according to the increase in spot trading volume to obtain the second clearing function.

[0373] Based on the determined long-term decrease in trading volume, the proportion of spot trading volume in the first clearing function is increased accordingly to obtain the spot trading volume value. Therefore, the first spot trading volume in the first clearing function can be adjusted using the increase in spot trading volume to obtain the second clearing function.

[0374] This application provides a grid clearing method. First, based on the power generation characteristics of each associated power plant, low-level data abstraction is performed to quantify various factors affecting grid clearing into a first data vector, thereby reducing the data complexity of the power generation characteristics. Then, long-term trading prices are predicted based on the first data vector and a preset long-term electricity price function library. This predicts the corresponding long-term trading prices based on the actual power generation characteristics of each associated power plant and establishes a corresponding long-term trading stack. The accuracy of grid clearing is improved by ensuring the accuracy of the long-term trading prices. Further, a spot trading stack is established based on the spot trading prices of each associated power plant, and a first clearing function is determined through the long-term trading stack and the spot trading stack. Correspondingly, based on the determined first clearing function, the power supply and demand balance state of the target grid at the time of clearing under the first clearing function is determined, and the first clearing function is adjusted in real time according to the actual power supply and demand balance state of the target grid to obtain a second clearing function. This allows for dynamic adjustment of the target grid's clearing function based on its supply and demand status. A specific clearing function is established by constructing a trading stack in real-time using long-term and spot trading prices. This ensures that grid clearing effectively considers the impact of power generation characteristics on associated power plants, as well as changes in spot trading prices for each associated power plant, thereby determining a more accurate clearing function and improving grid clearing accuracy. In this scheme, to better illustrate its data-driven approach, measured data is used to fit the power forecast curve for scheduling. The clearing agent quantifies electricity price and quantity, as well as long-term and short-term prices, allowing the algorithm within the data and information flow to be clearly defined. This agent needs to be constructed to determine and clear electricity prices. The algorithm, i.e., a neural network, needs to be selected based on the characteristics of the data and information flow.

[0375] Example 3

[0376] In Example 3, the power function requirement for grid-side dispatch control over power plant is specifically peak-shaving dispatch in a high-frequency load change scenario for thermal power plants. The functional operator is a thermal power peak-shaving dispatch operator, which is used to implement peak-shaving dispatch in a high-frequency load change scenario for thermal power plants. Example 3 describes the process of using an intelligent agent built based on the aforementioned thermal power peak-shaving dispatch operator in a digital power system.

[0377] In scenarios where thermal power generation and renewable energy generation coexist, the instability of renewable energy generation leads to frequent load fluctuations in thermal power generation. Under these high-frequency load fluctuations in thermal power scenarios, the current lack of a corresponding digital cost-calculating peak-shaving dispatch scheme makes it difficult to implement safe and stable peak-shaving dispatch, potentially impacting the lifespan of the power system. Ensuring the lifespan of the power system is a critical issue. In light of this, the inventors propose a peak-shaving dispatch method for thermal power in high-frequency load fluctuation scenarios using a digital power system based on a multi-functional intelligent agent and a thermal power peak-shaving dispatch operator. Based on various data feedback from thermal power plants, this method comprehensively considers the load capacity, performance, cost of thermal power units, and changes in thermal power load demand. By employing digital technology, it analyzes the relationship between power plant costs and thermal power load changes, thereby ensuring the safety and stability of peak-shaving dispatch.

[0378] Figure 5A illustrates a power supply scenario. As shown in Figure 5A, this scenario involves multiple power generation methods, including thermal power generation and photovoltaic (PV) power generation. PV power generation, as a new energy source, is gradually increasing its share in the energy consumption structure due to its zero emissions, zero pollution, and sustainability. However, compared to thermal power generation, PV power generation exhibits greater instability. Therefore, in the power supply scenario shown in Figure 5A, thermal power plants that generate electricity need to frequently change their load. In practice, thermal power plants need to cooperate with the grid's peak-shaving dispatch. Taking Figure 5A as an example, the left side of Figure 5A shows three thermal power plants, each capable of storing and releasing electrical energy. The released electrical energy from the thermal power plants is transmitted to the grid through their own electrical connections, ultimately serving the thousands of households that purchase electricity from the grid.

[0379] Each thermal power plant typically has multiple generating units, and the equipment performance, supporting infrastructure, and maintenance costs often vary between plants. Furthermore, the equipment durability, safety features, ramp-up capabilities for peak-shaving scheduling, and power generation costs also differ among the units. For thermal power plants, cooperating with peak-shaving scheduling requires consideration of whether it will lead to operational losses; additionally, it needs to determine whether the plant itself is capable of operating under peak-shaving conditions. On the grid side, it's necessary to consider not only the pricing of each unit within the thermal power plant and their individual capabilities and performance, but also various scheduling factors to ensure the safe and stable implementation of peak-shaving scheduling.

[0380] Figure 5B is a flowchart of a peak-shaving dispatching method for a thermal power plant under a high-frequency load change scenario, provided in an embodiment of this application. Figure 5C is a signaling interaction diagram of a peak-shaving dispatching method for a thermal power plant under a high-frequency load change scenario. In Figure 5B, the implementation of the scheme is mainly introduced from the perspective of the thermal power plant. In Figure 5C, the implementation of the scheme is introduced interactively from both sides (power plant side and grid side).

[0381] As shown in Figure 5B, an embodiment of this application provides a peak-shaving dispatching method for thermal power plants under high-frequency load changes, including:

[0382] S501. Calculate the heat storage capacity of energy storage devices in a thermal power plant.

[0383] Each thermal power plant is equipped with energy storage devices. These devices store the heat generated by the coal combustion in the thermal power units and release it when needed, converting the heat into electrical energy which is then supplied to the power grid. In one possible implementation, the energy storage devices within the thermal power plant include, but are not limited to, boilers, turbine systems, heaters, and deaerators.

[0384] The heat storage margin of each energy storage device is divided into two different types: a heat storage margin characterization value and a heat release capacity characterization value. The heat storage margin characterization value indicates the energy storage device's ability to store further heat. The heat release capacity characterization value indicates the energy storage device's ability to release further heat. The heat storage margin characterization value can be expressed as a specific heat value or as a percentage compared to the rated stored heat. Similarly, the heat release capacity characterization value can be expressed as a specific heat value or as a percentage compared to the pre-rated stored heat. Both heat storage margin and heat release capacity are related to the size and performance of the unit.

[0385] The heat storage capacity of energy storage devices within a thermal power plant reflects the plant's current output capacity from both "storage" and "retrieval" perspectives. In practice, based on the actual needs of the power grid, the heat storage capacity can be statistically analyzed using two different dimensions. For example, it can be analyzed according to the dimensions of "day-ahead" and "intra-day." The thermal power plant provides the day-ahead heat storage capacity of its energy storage devices, allowing the power grid to prepare in advance and conduct orderly peak-shaving scheduling based on this data. The thermal power plant provides the intra-day heat storage capacity of its energy storage devices, enabling the power grid to achieve accurate, flexible, and reliable peak-shaving scheduling based on highly real-time data.

[0386] In one possible implementation, the heat storage margin sent by the thermal power plant to the grid can be specifically refined to the heat storage margin corresponding to each thermal power unit of the plant. It should be noted that the heat storage margin can be calculated by the thermal power plant based on the coal type parameters and thermal system design of the selected target coal type. The choice of coal type affects heat consumption and coal consumption. Table 2 is a data table of a digital coal type selection model. Table 2 shows the testing items for various coal types such as GEB 5200, GEB 4800, and GEB4700. Here, the testing items can also be considered as the coal type parameters mentioned above. These coal type parameters include, but are not limited to, total moisture, air-dried basis moisture, as-received ash, dry ash-free basis powdering, as-received carbon, as-received hydrogen, total sulfur, and lower heating value. In Table 2, LHV represents the lower heating value of the coal type, and HHV represents the higher heating value of the coal type.

[0387] Table 2

[0388] The digital coal selection model can select coal types by extracting features from the data shown in Table 2. Because the hydrogen (H) portion of carbon combines with oxygen to produce water during physical combustion in a boiler, this water absorbs a large amount of heat, known as the "latent heat of vaporization," making it a crucial factor in coal selection. This hydrogen content is measured by the percentage difference between its HHV and LHV values, as shown in the last row of Table 2. The smaller this value, the less heat is lost due to latent heat of vaporization, resulting in more heat generated per unit of coal for power generation and better economic benefits. In practical applications, the digital coal selection model obtains the percentage difference between the high and low calorific values ​​of each candidate coal type (this value can be understood as a low-order abstract feature) based on their respective high and low calorific values. The coal type with the smallest percentage difference between its high and low calorific values ​​among the candidate coal types is then identified as the target coal type. Taking Table 2 as an example, coal type GEB 5200 has the smallest percentage difference between its higher and lower heating values, specifically 5.90%. Therefore, GEB 5200 can be selected as the target coal type. By selecting the coal type with the smallest percentage difference between its higher and lower heating values ​​as the target coal type, the heat generated per unit of coal can be maximized.

[0389] The formulas for calculating the higher heating value (HHV) and lower heating value (LHV) of coal are as follows: HHV = LHV + 25 * M + 25 * 9 * H

[0390] M represents the total water content of coal, and H represents the basic hydrogen content of coal. In the above formulas, the units for HHV and LHV are both kcal / kg. In one example, M = 23 and H = 4.21. It should be noted that in the above formulas, when converting between the calorific unit kcal and kJ, the values ​​of M and H must be converted according to the conversion factors for different calorific units.

[0391] S502. Send the correlation data between the load, performance and cost of each thermal power unit in the thermal power plant, as well as the heat storage margin, to the grid side so that the grid side can determine the target units for expected auxiliary peak shaving and issue peak shaving dispatch instructions based on the changes in thermal power load demand, the correlation data provided by each thermal power plant and the heat storage margin.

[0392] To facilitate the grid's access to, analysis of, and peak-shaving scheduling of data from each thermal power plant and unit, this application requires each thermal power plant to provide its own data to the grid. Specifically, the thermal power plant sends the grid with correlation data between the load, performance, and cost of each unit, along with the heat storage margin calculated in step S501. It should be noted that in this embodiment, the correlation data sent by the thermal power plant to the grid is also calculated based on the coal type parameters of the selected target coal type.

[0393] Table 3 shows the correlation parameters between heat consumption, coal consumption, and electricity price under different loads. Column 1 of Table 3 displays various power plant loads (e.g., 100% load, 95% load, 65% load, 45% load, etc.). Columns 2 to 24 of Table 3 show various heat consumption, coal consumption, and system parameters under different loads. As shown in Table 3, under high-frequency load changes, the actual plant power consumption rate (column 7 of Table 3) of thermal power plants constantly changes. For example, under 100% plant load, the actual plant power consumption rate is 5.71%, and under 45% plant load, the actual plant power consumption rate is 7.37%.

[0394] By reporting the data shown in Table 3 to the grid, the grid possesses the load conditions and thermal energy status of each thermal power plant. This allows for better data classification and aggregation, enabling the allocation of different weights and priorities to units based on their operating conditions and thermal energy reserves. Since the calculations in this process are influenced by multiple variables, the features extracted from the data shown in Table 3 can be called intermediate-order data abstract features (or simply intermediate-order features). The chain rule is used to transmit these intermediate-order data abstract features, reflecting the power plant's energy consumption and thermal energy reserve levels, thus assisting in peak-shaving scheduling in high-frequency load changes at thermal power plants.

[0395] Table 3

[0396] Table 4 shows the electricity price and cost data for a selected target coal type under different operating conditions (power plant load). The grid subsidy coefficient per kilowatt-hour is the electricity price supplied by the grid (excluding limestone costs), which is the fuel portion of the capacity price and generally accounts for more than 50% of the total electricity price.

[0397] The following is an example formula for calculating the electricity grid subsidy coefficient per kilowatt-hour: CERm=U*N / E(1-AUX)

[0398] Where U represents the coal price, which is also the fuel cost subsidy, in US dollars per ton (USD / t). For example, U = 69 USD / t. E represents the unit's heat consumption rate, i.e., the heat consumption per kilowatt-hour of electricity, in kcal / kWh. N represents the specific heat index, specifically the heat required per kilowatt-hour of electricity, in kcal / kWh. AUX represents the plant power consumption rate, in %.

[0399] In Table 4, SHRW represents the weighted specific heat index, in kJ / (kW·h). Ea represents the electricity supplied (or generated), in kW·h. Ep represents the electricity cost, in USD.

[0400] SHRcc is the specific heat index, measured in kJ / (kW·h). U is the coal price, measured in USD / t. ECRm is the electricity cost per kilowatt-hour, measured in USD / (kW·h). ECRm represents the electricity charge ratio, i.e., the cost of coal for producing one kilowatt-hour of electricity, measured in USD / (kW·h). It should be noted that if the daily load is evenly distributed throughout the month, the weighted specific heat index equals the weighted index, i.e., SHRW = SHRcc.

[0401] The formula for calculating ECRm is: ECRm=SHRcc×(1 / HHV)×U / 1000

[0402] The formula for calculating Ep is: Ep = NEOm * CERx

[0403] In the formula above, NEOm represents net electricity generation in kW·h. CERx represents the electricity price in USD / (kW·h).

[0404] As shown in Table 4, by analyzing the correlation between electricity prices and load at different values, the cost of fuel reflected in the fuel of each thermal power plant in the power grid can be obtained. This allows the power grid to extract the unit performance, environmental impact, and fuel cost of power plants under different load conditions. The extracted features, influenced by multiple variables (including coal cost), are called high-order abstract features. These linear high-order features, generated by integrating coal cost and other factors based on mid-order features, encompass the correlation between fuel, electricity consumption, and electricity prices.

[0405] In other words, in the technical solution of this application, the grid side determines the target units for expected auxiliary peak shaving and issues peak shaving dispatch instructions based on changes in thermal power load demand, the correlation data provided by each thermal power plant, and the heat storage margin, including:

[0406] Based on the associated data provided by each thermal power plant and the heat storage margin, the power grid side extracts intermediate-level features, which include features reflecting the power plant's energy consumption level and heat energy reserve level. The power grid side integrates the coal price cost in the associated data and the intermediate-level features to obtain high-level features, which include the relationship between fuel, electricity, and electricity price.

[0407] Based on the changes in thermal power load demand, the intermediate-order characteristics, and the high-order characteristics, the power grid side determines the target generating units for expected auxiliary peak shaving and issues peak shaving dispatch instructions.

[0408] As shown in Table 4, the higher the load, the smaller the grid subsidy coefficient per kilowatt-hour, indicating lower fuel costs. Lower fuel costs under the same operating conditions indicate better unit performance and lower coal consumption. Therefore, the grid can compare the fuel costs of different thermal power units under the same operating conditions (e.g., the grid subsidy coefficient per kilowatt-hour shown in Table 4) to determine the performance differences between them. For example, thermal power unit A has a grid subsidy coefficient of 0.0311 USD / (kW·h) at 100% load; thermal power unit B has a grid subsidy coefficient of 0.0319 USD / (kW·h) at 100% load, and both units use the same type of coal. By comparing this coefficient, it can be determined that thermal power unit A performs better than thermal power unit B.

[0409] Table 4

[0410] Based on the data reported by each thermal power plant (including correlation data between the load, performance, and cost of each thermal power unit within the plant, as well as heat storage margins, etc., where the correlation data can be referred to in the examples in Tables 3 and 4), the grid side can have a full understanding of the thermal power units of each plant. Based on this, the grid side can perform peak-shaving dispatch functions. In specific implementation, the grid side can determine the target units for expected auxiliary peak-shaving based on changes in thermal power load demand, correlation data provided by each thermal power plant, and heat storage margins, and then issue peak-shaving dispatch instructions to the target units.

[0411] Before the grid side determines the target generating units for the expected auxiliary peak-shaving dispatch, it can issue peak-shaving load curves for each thermal power plant's generating units. In one example implementation, the peak-shaving load curve is two-dimensional, with the horizontal axis representing time and the vertical axis representing power. The peak-shaving load curves may differ for different thermal power units. Essentially, the peak-shaving load curve expresses the load requirements for the thermal power units.

[0412] Thermal power plants can determine which units meet the peak load requirements based on the peak load curves of their respective units, and calculate pricing information for providing auxiliary peak load services to these units. Since cooperating with peak load scheduling is an additional auxiliary task provided by thermal power units, a certain fee needs to be charged to the grid. The fee is presented in the form of a price quote. Thermal power plants calculate pricing information for providing auxiliary peak load services by combining the performance of the thermal power units, the requirements of the peak load curves, the type of coal burned, and fuel information such as coal consumption, heat consumption, fuel costs, and equipment costs. This pricing information is sent to the grid, allowing the grid to compare and determine the thermal power units expected to provide auxiliary peak load services; these units are called target units. The number of target units within a thermal power plant is variable. For example, thermal power plant A may have 3 target units; thermal power plant B may have 1 target unit; and thermal power plant C may have no target units identified. Combining Tables 3 and 4, the grid side can use data-driven and deep learning capabilities to determine the target units for expected auxiliary peak-shaving services based on changes in thermal power load demand, related data provided by each thermal power plant, heat storage capacity, and quotation information reported by each thermal power plant, by balancing costs and indicators through data-driven approaches.

[0413] For example:

[0414] Each thermal power plant provides its own data (Table 3) to the power grid. The power grid then compares and selects units based on the data uploaded by each plant. For example, if the power grid learns that Plant A is operating at 95% load and Plant B at 60% load, and that Plant A and Plant B have corresponding heat release capacity values ​​of 70% and 80% respectively, then the grid needs to use its deep learning capabilities to determine whether to prioritize heat release capacity or load margin under this operating condition when issuing dispatch instructions to select the appropriate units for grid connection. This selection strategy is based on data-driven and deep learning-driven historical data to prevent system safety hazards.

[0415] In one possible implementation, the power grid can establish a set of typical load curves based on load fluctuation data of thermal power units under various operating conditions, as well as the correlation data between the load, performance, and cost of each thermal power unit. The peak-shaving load curves for thermal power units on the power grid side are selected from this pre-established set of typical load curves. The thermal power plant calculates the pricing information for providing auxiliary peak-shaving services to thermal power units that meet the requirements of the corresponding peak-shaving load curves, based on the peak-shaving load curves and power supply quality of each thermal power unit. Considering power supply quality while referencing the peak-shaving load curves ensures that the pricing matches the power supply quality, avoiding the problem of inflated pricing with poor power supply quality.

[0416] Figure 5D is a flowchart illustrating the implementation of determining the target unit for expected auxiliary peak shaving according to an embodiment of this application. Referring to Figure 5D, to determine the target unit for expected auxiliary peak shaving, the technical solution of this application can combine the ramp-up coefficients and pricing information of each thermal power unit, using data-driven methods to balance cost and performance indicators, and select the thermal power unit with a suitable price from multiple expected auxiliary peak shaving thermal power units as the target unit. As shown in Figure 5D, this process includes:

[0417] S5021. Based on the changes in thermal power load demand, the associated data provided by each thermal power plant, and the heat storage margin, the power grid side generates peak load curves for each thermal power unit and calculates the ramp-up coefficient for each thermal power unit.

[0418] Understandably, during thermal power generation, the demand for thermal power load may fluctuate frequently due to the parallel operation of thermal power generation and renewable energy generation. For example, the demand for thermal power load is lower at midday and higher in the evening; it is higher during cloudy or rainy periods and lower during sunny periods. By combining the correlation data on load, performance, and cost provided by each thermal power plant, artificial intelligence algorithms can be used to analyze and generate peak-shaving load curves for each thermal power unit through feature extraction and other methods. This peak-shaving load curve relies on the extraction and analysis of multi-level and multi-tiered data to realize a data-driven peak-shaving scheduling scheme, which is specifically displayed by the generated peak-shaving load curve.

[0419] Furthermore, in the technical solution of this application, in order to select the target unit, the grid side can first calculate the ramp rate of each thermal power unit.

[0420] In this application's technical solution, the ramp rate coefficient is a parameter used to measure the performance of a thermal power unit. A higher ramp rate coefficient indicates better performance (ramp capability) of the thermal power unit, enabling it to better perform auxiliary peak-shaving tasks. The following provides an exemplary method for calculating the ramp rate coefficient of a thermal power unit.

[0421] To calculate the ramp-up coefficient of a specific thermal power unit, the grid side can extract the ramp-up rate, steam pressure, and rated generating capacity from the correlation data between the unit's load, performance, and cost. Then, based on the unit's ramp-up rate, steam pressure, rated generating capacity, heat storage margin, and the required ramp-up time from the power plant, the ramp-up coefficient is calculated. Below is an exemplary formula for calculating the ramp-up coefficient of a thermal power unit: P = S * T * Q * C / R

[0422] In the formula, P is the ramp coefficient, S is the ramp speed, T is the ramp time required by the power grid, Q is the steam pressure, C is the heat storage margin, and R is the rated generating capacity. Table 5 exemplarily shows the operating conditions, ramp speed, grid-required ramp time, steam pressure, heat storage margin, and calculated ramp coefficients of three different thermal power units (W1, W2, and W3). The steam pressure Q is expressed as a percentage of the rated pressure. Combining the above formula, the calculation method for the ramp coefficient of the thermal power unit shown in this formula is as follows: calculate the product of the ramp speed, ramp time, steam pressure, and heat storage margin of the thermal power unit; calculate the ratio of the product to the rated generating capacity of the thermal power unit, and use this ratio as the ramp coefficient of the thermal power unit. The calculation of the ramp coefficient allows for the quantitative measurement of the capacity level of each thermal power unit and enables horizontal comparison between them. The data-driven quantitative comparison of thermal power units enriches the theoretical basis of peak-shaving scheduling from a data perspective, and helps to achieve more reliable and orderly high-frequency peak-shaving scheduling.

[0423] Taking Table 5 as an example, the heat storage margin is represented by C. As one of the parameters for calculating the ramp-up coefficient of thermal power units, the C value can specifically represent the heat release capacity characterization value in the heat storage margin. As shown in Table 5, the C values ​​for units W1, W2, and W3 are 90%, 50%, and 20%, respectively. This means that: the heat release capacity characterization value of unit W1 is 90%, and the heat storage margin characterization value is 10%; the heat release capacity characterization value of unit W2 is 50%, and the heat storage margin characterization value is 50%; the heat release capacity characterization value of unit W3 is 20%, and the heat storage margin characterization value is 80%. In other words, for the same unit, the sum of its heat release capacity characterization value and its heat storage margin characterization value is 100%.

[0424] Table 5

[0425] S5022. Based on the ramp coefficient of each thermal power unit, several thermal power units expected to assist in peak shaving are preliminarily determined.

[0426] In this embodiment, the power grid can initially determine multiple units based on the relative magnitudes of their ramp coefficients, with the intention of having these units assist in peak shaving. Referring to the last column of Table 5, thermal power units with different ramp coefficients can be rated (peak shaving dispatch level) according to their ramp coefficients. For example, thermal power units with a ramp coefficient greater than or equal to 0.8 are classified as Level A; those with a ramp coefficient less than 0.8 but greater than or equal to 0.5 are classified as Level B; and those with a ramp coefficient less than 0.5 are classified as Level C. This establishes a correspondence between thermal power units with different ramp coefficients and peak shaving dispatch levels.

[0427] In practice, based on the peak-shaving dispatch level corresponding to each thermal power unit, several thermal power units expected to assist in peak shaving are initially determined. For example, the grid side can prioritize Class A thermal power units. If Class A thermal power units are still insufficient for dispatch, Class B thermal power units can be further selected as expected auxiliary peak-shaving units.

[0428] S5023. Determine whether there are any thermal power units among the multiple expected auxiliary peak-shaving units whose price information is less than or equal to K times the threshold price. If yes, proceed to S5024; otherwise, proceed to S5025.

[0429] S5024. This thermal power unit has been identified as the target unit for expected auxiliary peak shaving.

[0430] S5025. This thermal power unit will not be selected as the target unit for expected auxiliary peak shaving.

[0431] Understandably, each thermal power unit, if required to cooperate with auxiliary peak-shaving services, may incur heat and energy consumption beyond its expected power generation targets, resulting in certain cost expenditures. Therefore, thermal power units can calculate their bids and submit them to the grid. The grid compares the bid information submitted by each thermal power unit with K times the threshold bid. If the bid is less than or equal to K times the threshold bid, the grid considers the bid relatively reasonable and acceptable, as shown in step S5024, and such units can be designated as target units for auxiliary peak-shaving. Conversely, if the bid is higher than K times the threshold bid, the grid considers the bid too high and will no longer prioritize such units as target units.

[0432] The threshold price is a reference value obtained from historical price information obtained through deep learning on the grid side. In practical applications, the grid side can comprehensively learn historical price information under high-frequency load changes in thermal power, combine this historical price information with the performance and cost of the thermal power units themselves, and use data-driven analysis to weigh costs and indicators to determine a reasonable reference price. K is a coefficient greater than 1, and in one example, K takes a value between 1.1 and 1.3. For example, if K is 1.3, it means that if the price information of a thermal power unit exceeds the threshold price by more than 30%, it is considered unsuitable as a target unit. Conversely, if the price information of a thermal power unit exceeds the threshold price by no more than 30%, it is considered suitable as a target unit.

[0433] For a target generating unit, the grid side can send peak-shaving dispatch instructions to the thermal power units to which the unit belongs.

[0434] S503: Receive peak-shaving dispatch instructions issued by the power grid side.

[0435] Peak-shaving dispatch instructions carry the unit identifier of the target unit and the peak-shaving requirement information for that unit. For example, if a peak-shaving dispatch instruction carries the unit identifier W1, the thermal power plant receiving the instruction can know that it is issued to the thermal power unit with the unit identifier W1. If a thermal power plant contains multiple target units, the grid side can send peak-shaving dispatch instructions to each thermal power unit in parallel. This group-sending dispatch improves peak-shaving efficiency. Because the peak-shaving dispatch instruction carries the unit identifier, it allows the thermal power plant to accurately forward peak-shaving requirement information.

[0436] S504. Execute auxiliary peak shaving services based on peak shaving requirement information.

[0437] As mentioned earlier, peak-shaving dispatch instructions carry peak-shaving requirement information for the target generating units. Specifically, this peak-shaving requirement information includes the peak-shaving load curve. In one possible implementation, the peak-shaving requirement information also includes ramp-up speed requirement information. The peak-shaving load curve sets requirements for the target generating units from a load perspective, while the ramp-up speed requirement information sets requirements for the target generating units from a ramp-up speed perspective. When the target generating unit receives the peak-shaving requirement information, it needs to make a judgment based on its own actual situation to confirm whether it can meet the above requirements.

[0438] Figure 5E is a flowchart of a fault diagnosis and operating condition diagnosis method provided in an embodiment of this application. The following, with reference to Figure 5E, exemplarily describes one implementation of this step.

[0439] As shown in Figure 5E, after identifying the peak-shaving requirement information in the peak-shaving dispatch instruction, the target unit of the thermal power plant first determines whether it has any equipment or communication faults, and then determines whether the target unit meets the ramp-up speed requirement information. Specifically, as shown in Figure 5E, it first determines whether there are any equipment or communication faults. After this determination, assuming there are no equipment or communication faults, it further determines whether the ramp-up speed requirement information is met. If there are no equipment or communication faults in the target unit (i.e., neither equipment nor communication faults exist), and the target unit meets the ramp-up speed requirement information, then auxiliary peak-shaving services are executed according to the peak load curve and ramp-up speed requirement information in the peak-shaving requirement information. However, if there are equipment or communication faults in the target unit, it means that the target unit cannot cooperate in completing the auxiliary peak-shaving service due to hardware or communication conditions. In this case, the thermal power plant needs to report the fault message of the thermal power unit to the grid side, so that the grid side can be informed of the fault status of the target unit in a timely manner. If the target unit does not meet the required ramp speed, a notification message indicating that the operating condition is not met is sent to the power grid side.

[0440] It should be noted that, in this embodiment, the power grid can adjust peak-shaving dispatch instructions in response to feedback from thermal power plants indicating unmet operating conditions. For example, based on the unmet operating condition feedback, the peak-shaving load curve in the peak-shaving requirement information can be adjusted, and / or, the value in the ramp-up speed requirement information can be adjusted. By adjusting the peak-shaving load curve (e.g., reducing the vertical axis value of the load curve) or reducing the value in the ramp-up speed requirement information, the threshold for the target units can be lowered, enabling more units to become target units for peak-shaving dispatch.

[0441] Furthermore, in this embodiment, the grid side can also relearn and raise the threshold price based on the feedback from each thermal power plant indicating that the operating conditions are not met, thus allowing more units to be considered as target units for expected auxiliary peak shaving. Including more units in the queue of target units for expected auxiliary peak shaving expands the range of selected units and avoids the problem of insufficient capacity to provide auxiliary peak shaving services among units with suitable prices. Including more units in the target unit queue allows more units to participate in auxiliary peak shaving work and avoids the problem of high-performance thermal power units being unable to contribute significantly to auxiliary peak shaving due to price constraints. Continuous deep learning of the threshold price ensures that it can be updated in a timely manner based on the status and prices of thermal power units, making peak shaving scheduling dynamic and flexible.

[0442] As mentioned earlier, in this application, thermal power units, especially the selected target units, should cooperate with grid-side dispatching to assist in peak shaving. The peak shaving dispatching method for high-frequency load changes in thermal power provided in this application, after initially identifying multiple expected auxiliary peak shaving thermal power units on the grid side, further includes:

[0443] The power grid sends heat pre-dispatch instructions to the multiple thermal power units expected to assist in peak shaving. The purpose of issuing heat pre-dispatch instructions is to enable each thermal power unit expected to assist in peak shaving to prepare its heat capacity for the upcoming load adjustment work. Specifically, heat pre-dispatch may require energy storage devices to store more heat or to release more heat. The heat pre-dispatch instructions can be issued by the power grid to the thermal power units after the power grid has learned of their heat storage capacity. Given the heat storage capacity, the power grid can ensure that the content of the issued heat pre-dispatch instructions and the heat adjustment range are more closely matched to the performance and capacity of the thermal power units when performing peak shaving dispatch and issuing heat pre-dispatch instructions.

[0444] As mentioned earlier, the grid side determines the target generating units for auxiliary peak shaving based on bidding information. For target generating units, it is assumed that the thermal power plant and the grid side have reached an agreement on the bidding. The target generating units selected by the grid side, upon receiving a heat pre-dispatch instruction, will coordinate with the heat storage devices within their respective thermal power plants to execute the heat pre-dispatch. For example, if the heat pre-dispatch instruction requires the target generating unit to store an additional 12% of heat, the target generating unit will store an additional 12% of heat according to the instruction. Thus, when auxiliary peak shaving services are needed, the corresponding load adjustment can be completed using the pre-stored heat.

[0445] Figure 5C illustrates the peak-shaving dispatching process under high-frequency load changes in thermal power, showcasing the interaction content and timing from both the power plant and grid sides. As shown in Figure 5C, the power plant selects the coal type and calculates the heat storage margin. The calculated heat storage margin, along with related data (specifically, the correlation data between the load, performance, and cost of each thermal power unit within the power plant), is sent to the grid side. The specific types of related data can be found in Tables 3 and 4. The grid side digitally processes the data received from each power plant, performing multi-level abstraction, extraction, and analysis, and using artificial intelligence techniques to capture the relationships between the data. Then, based on the above extraction and analysis, the grid side generates peak-shaving load curves for each thermal power unit and sends the curves to the corresponding power plants. Additionally, heat pre-dispatch instructions can also be issued. Based on the actual situation of each thermal power unit, the power plant calculates bidding information to meet the requirements of the peak-shaving load curves to assist in peak-shaving, and reports the bidding information to the grid side. At this point, the grid can obtain the corresponding price information for each thermal power unit of each thermal power plant, and make horizontal comparisons based on the price information and the calculated ramp coefficients of the thermal power units. Through data-driven analysis, it balances costs and indicators to select the target unit. The grid then issues peak-shaving dispatch instructions to the thermal power plants, and the target units cooperate to perform heat pre-dispatch and auxiliary peak-shaving services based on the peak-shaving requirements information in the peak-shaving dispatch instructions.

[0446] The technical solution of this application has the following outstanding advantages:

[0447] With the increasing proportion of renewable energy generation in the power system, thermal power plants, from design to operation, must pay attention to storing heat energy to meet the grid's auxiliary peak-shaving service needs. The pricing of auxiliary peak-shaving services needs to be market-driven and collaborative. Therefore, the technical solution proposed in this application, through the aforementioned method, can meet the requirements of marketization and collaboration.

[0448] Furthermore, the technical solution of this application introduces data analysis into the market-based dispatching of high-frequency load changes in thermal power plants. Through quantified tables 2 to 4, the grid side (combined with tables 3 to 4) can achieve digital dispatching, and the power plant side (combined with table 2) can achieve digital coal preparation. The key lies in feature extraction based on the data. The data-driven dispatching process makes peak-shaving dispatching more standardized and digitalized, resulting in better coordination and clearer cost information. The grid can prioritize and rank units based on cost and performance indicators, thereby using data as a theoretical basis to better select suitable units for peak-shaving, leading to more stable, safe, and ideal peak-shaving results.

[0449] Furthermore, the technical solution in this application simultaneously considers grid security factors such as ramp rate and peak load curves. These key data points can guide power quality pricing, resulting in good economic efficiency. The pricing of high-frequency peak-shaving dispatch and auxiliary peak-shaving services takes into account the costs of the grid and power plants under market conditions.

[0450] This embodiment collects and processes the natural production factors of thermal power using digital tools, abstracting their data characteristics. Then, based on the thermal power electricity price formula, it obtains a metering model for the heat consumption, coal consumption, and electricity costs of thermal power under full load conditions. Through digital research on this model, data-driven digital logic can be applied to the power industry, enabling the digitalization of thermal power dispatch and operation. Simultaneously, by studying the control production factors on the grid side, such as electricity prices, it establishes a correlation between thermal power production factors and electricity prices and other complex production factors. By setting low-order, mid-order, and high-order data-abstracted production factors, intermediate results or production factors may not possess real-world physical meaning, such as sub-frequency or low-voltage overexcitation. These data-abstracted production factors are the key to unlocking artificial intelligence (digital) power.

[0451] Figure 5F is a flowchart of another peak-shaving scheduling method for thermal power plants under high-frequency load changes, provided in an embodiment of this application. A brief description of Figure 5F follows.

[0452] As shown in Figure 5F, the thermal power plant first selects coal types based on their hydrogen content, then calculates the plant's heat storage capacity, and digitizes the correlation data between the load, performance, and cost of each thermal power unit within the plant. The power plant then sends this data to the grid, which uses the data to weigh costs and indicators to determine the target units for expected auxiliary peak shaving.

[0453] In practice, the grid side generates peak-shaving load curves for each thermal power unit based on changes in thermal power load demand and corresponding data provided by each thermal power plant, and then sends these curves to the thermal power units. It also calculates the ramp-up coefficient for each thermal power unit, initially identifying several units expected to assist in peak shaving, and sends pre-dispatch instructions for heat to these units. The thermal power plants calculate their bids for auxiliary peak-shaving services based on the peak-shaving load curves and send them to the grid side. The grid side then further considers cost factors and the performance data of the thermal power plants, weighing the bid information against a K-fold threshold bid to select target units. If all bids exceed the K-fold threshold bid, it indicates that the threshold bid is too low, meaning that deep learning of the threshold bid is needed to improve the pass rate of thermal power units meeting the load bid threshold.

[0454] The target generating unit coordinates with heat pre-scheduling, and after pre-scheduling is completed, it reports the heat storage to the grid again. The grid side determines whether there are equipment or communication faults in the target generating unit, and whether the ramp-up speed requirements are met. If equipment or communication faults exist, the grid needs to be notified; if the target generating unit does not meet the ramp-up speed requirements, the grid also needs to be notified. As shown by the dashed line in Figure 5F, for cases where the operating conditions are not met, measures such as relearning the threshold price can be taken, or the peak-shaving requirements in the peak-shaving dispatch instructions issued to the target generating unit can be adjusted.

[0455] Example 4

[0456] In Example 4, the power function requirement for realizing the grid-side dispatch control over the power plant side specifically refers to the water resource dispatch of cascade hydropower stations. The functional operator is the water resource dispatch operator, which is used to realize the water resource dispatch of cascade hydropower stations. Example 4 describes the process of using an intelligent agent built based on the aforementioned water resource dispatch operator in a digital power system.

[0457] There are a series of issues to consider in the water resource management of cascade hydropower stations. For example, a cascade hydropower station is generally composed of multiple single-stage power stations connected in series. The upstream and downstream power stations are connected through reservoir backwater, and the control range can extend from confluence to the entire basin. The risk of some uncontrolled events at the upstream power station can be transmitted through water flow fluctuations, superimposed and accumulated at the downstream power station, which may cause serious risk events. In addition, during the operation of each hydropower station in the cascade hydropower station, the safety of the reservoir (appropriate water level) and the safety of the generating units need to be considered. In addition, during the operation of each hydropower station in the cascade hydropower station, the safety of the generating units should be considered in the event of sudden grid accidents (such as failure to receive power from the interconnection point, sudden load shedding). Furthermore, under the premise of reservoir safety and the safety of hydropower generation equipment, how to improve the quality of power supply, promote the safe operation of the grid, increase power supply revenue, and achieve full utilization of water resources.

[0458] To address the aforementioned issues, this application proposes a water resource scheduling method for cascade hydropower stations based on a multifunctional intelligent agent-based digital power system and utilizing water resource scheduling operators. The method includes: acquiring water resource data for each hydropower station in the cascade system at the current moment; for each hydropower station, determining whether it is in a safe operating state based on its water resource data; when all hydropower stations are determined to be in a safe operating state, determining whether water resource scheduling should be performed based on the water resource data of each hydropower station, a preset electricity price scheme, and target historical data; if water resource scheduling is determined, then a water resource scheduling scheme is determined based on the water resource data of each hydropower station and the water resource scheduling model. Through the scheme in this application, when it is determined that each hydropower station is in a safe operating state, a water resource scheduling scheme that maximizes power generation revenue can be directly generated based on the water resource scheduling model trained from historical data and the current water resource data of each hydropower station, thus achieving rational water resource scheduling.

[0459] Figure 6A is a flowchart of a water resource scheduling method provided in an embodiment of this application. Referring to Figure 6A, the water resource scheduling method disclosed in this application includes:

[0460] S601: Obtain water resource data for each hydropower station in the cascade hydropower station at the current moment.

[0461] The water resource data for the hydropower station in this application includes, but is not limited to: the measured water level of the reservoir where the hydropower station is located (hereinafter referred to as "measured water level"), the measured rate of increase of the water flow velocity in the reservoir where the hydropower station is located (hereinafter referred to as "measured rate of increase of velocity"), the measured turbine static pressure (also known as "measured turbine hydrostatic pressure"), the rated speed of the turbine generator, the working power of the turbine generator, the installed capacity of the power station, the maximum / rated / minimum head, the maximum runaway speed, the rated output of the turbine and the diameter of the turbine tube, etc.

[0462] Table 6 shows the inflow of water into the reservoir of an upstream hydropower station as provided in this application. Table 6 presents the inflow of water into the reservoir of the upstream hydropower station from January to December for different years, with the unit being 10,000 m³. 3 .

[0463] For example, the following data can be read from Table 6: The inflow of water into the reservoir of the upstream hydropower station in January 2016 was 7.75 million m³. 3 In February 2023, the inflow of water into the upstream hydropower station was 12.57 million cubic meters. 3 .

[0464] It should be noted that upstream hydropower stations refer to hydropower stations located upstream in a cascade hydropower system.

[0465] Table 6

[0466] Table 7

[0467] Table 7 is a table showing the water volume changes of an upstream hydropower station provided in this application. Table 7 shows the daily water volume changes of the upstream hydropower station from January 23rd to January 30th. Specifically, it includes: current water storage (m³). 3 ), Current reservoir energy storage (100 million kWh), Current water level (m), Average inflow yesterday (m³) 3 / s), yesterday's average outflow (m³) 3 / s) and yesterday's power generation water consumption rate (m 3 / kWh).

[0468] S602, For each of the hydropower stations, determine whether the hydropower station is in a safe operating state based on the water resource data of the hydropower station.

[0469] The safe operating status in this application mainly refers to the water conservation safety status and the vibration safety status.

[0470] In this application, when determining whether a hydropower station is in a water conservation safety state, two indicators are mainly considered: the measured turbine hydrostatic pressure and the measured rate of velocity rise.

[0471] The measured turbine hydrostatic pressure, also known as the volute pressure, refers to the actual net water pressure at the inlet of the turbine volute. It is typically affected by water flow velocity, flow direction, turbine structure, and reservoir water level. The measured turbine hydrostatic pressure is crucial for assessing turbine operating efficiency, predicting turbine wear, and developing turbine maintenance plans.

[0472] The measured rate of rise refers to the actual speed at which the reservoir water level rises over a certain period of time, expressed as the change in water level per unit time. It is typically affected by inflow, outflow, rainfall, and the reservoir's regulation capacity. The measured rate of rise is of great significance for assessing the reservoir's storage capacity, predicting future water level changes, and formulating reservoir operation plans.

[0473] Typically, the measured rate of velocity rise and the measured turbine hydrostatic pressure can be obtained through the corresponding sensors; the theoretical rate of velocity rise and the theoretical turbine hydrostatic pressure can be obtained from the technical documents of the hydropower station.

[0474] It should be emphasized that the theoretical rate of rise in this application is the maximum actual rate of rise of the reservoir water level within a certain period of time during the safe operation of the hydropower station; the theoretical turbine hydrostatic pressure in this application is the maximum actual net water pressure at the turbine inlet during the safe operation of the hydropower station.

[0475] Table 8

[0476] Table 8 is a table of volute pressures measured on different dates according to this application. Table 8 shows the unit output (MW), upstream water level (m), downstream water level (m), volute pressure (MPa), and guide vane opening of different units detected at 09:04 AM, 15:00 PM, and 18:25 PM on August 18, 2018, and at 09:25 AM and 13:50 PM on August 19, 2018.

[0477] In one alternative implementation, the following steps can be used to determine whether each hydropower station in a cascade hydropower station is in a water conservation safe state:

[0478] For each hydropower station, the measured rate of velocity rise and the measured turbine hydrostatic pressure are obtained from the water resource data of that hydropower station. Then, the measured rate of velocity rise and the theoretical rate of velocity rise of that hydropower station are compared, and the measured turbine hydrostatic pressure of that hydropower station is compared with the theoretical turbine hydrostatic pressure.

[0479] If the measured rate of increase in velocity is less than the theoretical rate of increase in velocity, and the measured turbine hydrostatic pressure is less than the theoretical turbine hydrostatic pressure, then the hydropower station is determined to be in a water conservation safety state; if the measured rate of increase in velocity is greater than or equal to the theoretical rate of increase in velocity, or the measured turbine hydrostatic pressure is greater than or equal to the theoretical turbine hydrostatic pressure, then the hydropower station is determined not to be in a water conservation safety state.

[0480] It is understandable that equipment in hydropower stations, such as turbine generators, will generate vibrations during operation. The amplitude and frequency of these vibrations are important indicators for assessing the operating condition of the equipment. Excessive vibration may indicate wear, loosening, imbalance, or other potential problems in the equipment. In this application, the determination of whether a hydropower station is in a vibration-safe state mainly considers the actual operating power of the turbine engine and the vibration of its various components.

[0481] In one alternative implementation, it can be determined whether each hydropower station in a cascade hydropower station is in a vibration-safe state by the following method:

[0482] Step 1: For each hydropower station, obtain the power (actual operating power) of the turbine engine from the water resource data of that hydropower station, and record it as the first power.

[0483] It should be noted that during the process of the turbine engine's power increasing from 0 to its maximum power (full load power), within certain power ranges (referred to as the preset power range in this application), due to frequency resonance and other reasons, the vibration of various components of the turbine engine may be very large. To avoid damage to the turbine engine due to excessive vibration or vibration triggering the protection trip, this application focuses on the actual operating power of the turbine engine at the current moment (i.e., the first power in this application).

[0484] The second step is to determine whether the first power is within the preset power range.

[0485] It should be noted that the preset power range can be obtained from a variable power vibration test.

[0486] The third step is to quickly adjust the first power to the second power if the first power is within the preset power range; wherein the second power is outside the preset power range.

[0487] If the initial power is determined to be within a preset power range, it indicates that the vibration of various parts of the hydro-generator is currently large, which can easily cause damage to the hydro-generator. In this case, the operating power of the hydro-generator should be quickly adjusted. For example, the initial power can be quickly adjusted to a second power outside the preset power range to avoid excessive vibration of the hydro-generator due to frequency resonance or other reasons.

[0488] The fourth step is to determine whether the hydropower station is in a vibration safety state if the first power is outside the preset power range.

[0489] If it is determined that the first power is outside the preset power range, it can be determined that the turbine generator will not vibrate excessively due to frequency resonance or other reasons under this first power. At this point, it is necessary to further determine whether the hydropower station is in a vibration-safe state.

[0490] In one alternative implementation, the process of determining whether a hydropower station is in a vibration-safe state is as follows:

[0491] First, for each hydropower station, the rotational speed of the turbine engine and the actual vibration values ​​of different parts of the turbine engine are obtained from the water resource data of that hydropower station.

[0492] Then, after determining the actual turbine engine speed, the actual vibration values ​​of different parts of the turbine engine measured at that speed are compared with the corresponding allowable vibration values ​​for that part.

[0493] For each hydropower station, if the actual vibration values ​​of different parts of the turbine engine are all less than the corresponding permissible vibration values ​​at the current turbine engine speed, then the hydropower station is determined to be in a vibration-safe state; if the actual vibration value of a certain part of the turbine engine is greater than or equal to the permissible vibration value corresponding to that part, then the hydropower station is determined to be in a vibration-safe state and there is a vibration risk.

[0494] Table 9

[0495] Table 9 presents the permissible vibration values ​​for various parts of a hydro-generator provided in this application. As shown in Table 9, the hydro-generator has different permissible vibration values ​​at different speeds. For example, at speeds less than 100 r / min, the permissible vertical vibration value for the thrust bearing support of the vertical hydro-generator unit is 0.1 mm. If the actual measured vertical vibration value of the thrust bearing support at speeds less than 100 r / min is greater than 0.1 mm, then the vertical hydro-generator unit is considered to have a vibration risk.

[0496] Once it is determined that each hydropower station in the cascade hydropower station is in a state of water conservation safety and vibration safety, that is, when it is determined that each hydropower station is in a state of safe operation, the steps of the subsequent embodiments of this application can be continued.

[0497] It is understandable that if it is determined that each hydropower station in the cascade hydropower station is not in a safe operating state and there are water conservation safety issues or vibration safety risks, it is necessary to shut down for maintenance until the problems are resolved before restarting the power generation facilities in each hydropower station. After restarting, the steps of the subsequent embodiments of this application will be executed only after verifying that each hydropower station is in a safe operating state.

[0498] S603, when it is determined that all the hydropower stations are in the safe operation state, it is determined whether to carry out water resource scheduling based on the water resource data of each hydropower station, the preset electricity price scheme and the target historical data.

[0499] Understandably, hydropower generators need to precisely adjust the water level of the reservoir corresponding to the power station in order to meet the grid's electricity demand while achieving higher power generation revenue, in order to ensure the safe operation of the power plant's generating equipment.

[0500] Specifically, firstly, based on the water resource data and the preset electricity pricing scheme for each hydropower station, the actual water consumption for power generation and the actual power plant revenue for each station must be determined as the current actual data for the hydropower station. Then, based on the difference between the current actual data and the target historical data, it is determined whether water resource allocation is necessary. The target historical data includes the average water consumption for power generation and the average power plant revenue for the current month.

[0501] If the difference between the current actual data and the target historical data is too large, exceeding a preset difference threshold, then water resource allocation is determined to be necessary; otherwise, water resource allocation is determined not to be necessary. This application does not limit the specific value of the preset difference threshold; those skilled in the art can set the preset difference threshold according to actual needs.

[0502] The following embodiments of this application will describe in detail the process of acquiring current actual data and the process of acquiring target historical data.

[0503] In one alternative implementation, the following method can be used: based on the water resource data and preset electricity pricing scheme for each hydropower station, determine the actual water consumption for power generation and the actual power plant revenue for each hydropower station, and use this as the current actual data for the hydropower station. Specifically:

[0504] The first step is to obtain the actual power generation and actual water consumption for the target time period from the water resource data of each hydropower station. The target time period is a past period ending at the current time.

[0505] For example, the actual power generation and water consumption over the past hour can be obtained from the water resource data of hydropower station A at the current moment. The past hour is the target time period.

[0506] Both actual power generation and actual water consumption can be obtained using corresponding detection equipment. For example, the actual water consumption within a target time period can be measured using an ultrasonic flow meter.

[0507] The second step is to obtain the actual power generation and water consumption for each hydropower station based on the actual power generation and water consumption for the target time period.

[0508] For example, after obtaining the actual water consumption and actual power generation of hydropower station A in the most recent hour, the actual water consumption is divided by the actual power generation, and then divided by the target time period to obtain the actual power generation water consumption.

[0509] The actual water consumption for power generation in this application reflects the capacity characteristics of a hydropower station to require water resources per unit time and per unit of power generation. Actual water consumption for power generation is high-order abstract data, which can be compared with the specific heat index of thermal power plants and the wind resource utilization coefficient of wind power plants to determine the degree and performance of resource utilization by the power generation node; reflecting the specific function of the power station in terms of resource energy consumption.

[0510] The third step is to determine the actual power plant revenue for each hydropower station during the target time period based on the corresponding segmented electricity pricing scheme for the target time period and the actual water level of the hydropower station.

[0511] The segmented electricity pricing scheme (hereinafter referred to as the electricity pricing scheme) in this application is an electricity fee scheme related to the month and the reservoir water volume.

[0512] To facilitate understanding, we will take the segmented pricing scheme corresponding to a cascade hydropower station that includes an upstream power station (also known as an upstream reservoir power station) and a downstream power station (also known as a downstream reservoir power station) as an example to introduce the specific form of the segmented pricing scheme:

[0513] (a) Upstream reservoir power station:

[0514] Upstream reservoir power stations implement quarterly regulation, and pricing is based on different water levels in different seasons.

[0515] From January to March, the water level is between 250-260m above sea level. If the water level of the upstream reservoir is above 255m, the on-grid electricity price is J-0.001 / m; if the water level of the upstream reservoir is below 255m, the on-grid electricity price is J+0.001 / m. The electricity price calculation method for other water levels within the range is: if the water level is above the upper limit, the upper limit price is applied; if the water level is below the lower limit, the lower limit price is applied. Here, J in this application is the preset base value of the on-grid electricity price.

[0516] From April to June, the water level is between 210 and 250 meters above sea level. If the water level of the upstream reservoir is above 230 meters, the on-grid electricity price is J-0.001 / m; if the water level of the upstream reservoir is below 230 meters, the on-grid electricity price is J+0.001 / m. The electricity price for other water levels within the range is calculated as follows: if the water level is above the upper limit, the upper limit price applies; if the water level is below the lower limit, the lower limit price applies.

[0517] From July to September, the water level is between 250-260m above sea level. If the water level of the upstream reservoir is above 255m, the on-grid electricity price is J-0.001 / m; if the water level of the upstream reservoir is below 255m, the on-grid electricity price is J+0.001 / m. The electricity price for other water levels within the range is calculated as follows: if the water level is above the upper limit, the upper limit price applies; if the water level is below the lower limit, the lower limit price applies.

[0518] From October to December, the water level will be between 258 and 262 meters above sea level. If the water level of the upstream reservoir is above 260 meters, the on-grid electricity price will be J-0.001 / m; if the water level of the upstream reservoir is below 260 meters, the on-grid electricity price will be J+0.001 / m. The electricity price for other water levels within the range will be calculated as follows: if the water level is above the upper limit, the upper limit will be charged; if the water level is below the lower limit, the lower limit will be charged.

[0519] The water levels of the upstream and downstream reservoirs are between 262-265m and 115-120m, respectively. The electricity price is US$0.073 per kilowatt-hour. The water levels of the upstream and downstream reservoirs are considered safe when they are below 235m and 100m, respectively. During these periods, the reservoirs cannot generate electricity.

[0520] Table 10

[0521] Table 10 shows the relationship between reservoir water level and electricity price for an upstream hydropower station as provided in this application. The upstream reservoir power station can directly adjust the price based on the water level for each quarter, according to the information shown in Table 10, to obtain the actual electricity price PR corresponding to the current month. 实 .

[0522] In Table 10, water levels are represented by the letter L; the units for both water level and median water level are meters; the actual electricity price PR 实 The unit is (USD / degree). When the month is January to March, the water level is in the range of 250-260 meters, with a median water level of 255 meters. Within this monthly range, if the water level L is less than the median water level of 255 meters, then PR... 实 The values ​​are generated based on a full-load electricity price of 0.073 USD / kWh and a floating electricity price of 0.001. Table 10 shows the PR values ​​for each month when the water level L is at the median of the highest water level. 实 It is (0.073 + 0.001) (USD / degree); when the water level L is less than the median water level, PR 实 It is (0.073-0.001) (USD / degree).

[0523] This application analyzes water resource data and finds that the data for each quarter exhibits clustering characteristics; specifically, the quarterly water levels in Table 10 show segmented patterns, reflecting the hydrological and water level characteristics of the hydropower station. Only by better utilizing natural production factors (referred to as primary indicators in this application) can the benefits of water resource utilization be maximized, and only then can the market-based analysis of hydropower water costs be conducted. Therefore, this application adopts a method of targeting the median water level and comparing it with the median historical water level (which serves as one of the data samples for deep learning) to determine the measured inflow and the amount of water entering the reservoir, thereby determining a reasonable electricity generation price scheme.

[0524] It is understandable that if other patterns are discovered during the analysis of historical water resource data or other data, it may further guide the water level scheduling strategy adopted in this application, further improve the underlying logic of data-based scheduling agents emphasized in this application, and avoid errors caused by artificially abstracting data and specifying water resource scheduling schemes.

[0525] (b) Downstream reservoir power station:

[0526] Downstream reservoirs are subject to day-ahead scheduling, which is unrelated to the season.

[0527] If the water level is between 105-110m above sea level, and the water level of the downstream reservoir is above 107m, the on-grid electricity price is J-0.001 / m. If the water level of the downstream reservoir is below 107m, the electricity price for other water levels within the range is as follows: if the water level is above the upper limit, the price is based on the upper limit; if the water level is below the lower limit, the price is based on the lower limit.

[0528] Table 11 shows the relationship between reservoir water level and electricity price for a downstream hydropower station provided in this application. The downstream hydropower station can directly adjust the price based on the water level for each quarter, according to the information shown in Table 11, to obtain the actual electricity price (PR) for the current month. 实In Table 11, water levels are represented by the letter L. The units for both water level and median water level are meters. The actual electricity price PR... 实 The unit is (USD / degree). In Table 11, when the water level L is the median water level, PR... 实 It is (0.073 + 0.001) (USD / degree); when the water level L is less than the median water level, PR 实 It is (0.073-0.001) (USD / degree).

[0529] Table 11

[0530] For each hydropower station in a cascade hydropower station, after obtaining the corresponding segmented electricity pricing scheme for the target time period and the actual water level of the station, the actual power plant revenue for the target time period can be determined based on the following formula: The target formula is: Q3 = PR 实 ×P 实 (2.3)

[0531] The meanings of the letters in formulas (2.1), (2.2), and (2.3) are as follows:

[0532] Q1 represents the fee that the power grid should pay to the hydropower station when the power grid fails and the hydropower station cannot generate electricity normally (payment without negotiation); Q2 represents the actual revenue corresponding to the potential electrical energy at the current moment; Q3 represents the actual revenue corresponding to the actual generated electrical energy at the current moment. The sum of formulas (2.1) and (2.3) represents the actual power plant revenue of the hydropower station; the sum of formulas (2.1), (2.2), and (2.3) represents the water assets of the hydropower station.

[0533] D represents the number of hydroelectric generators included in the hydroelectric power station, and D is a positive integer greater than 1; PR 实 The current time represents the actual electricity price for the month; PR_flat represents the average electricity price within the target time period (i.e., the target pricing period); i is a turbine generator in a hydroelectric power station, i = 1, 2…D; P 实 is the power generation (actual electrical energy) within the target time period; Gi is the rated power of the i-th turbine generator in the hydropower station; Hi is the continuous power generation time of the i-th turbine generator in the hydropower station at its rated power within the target pricing time period; FiJ is the derated power of the i-th turbine generator in the hydropower station; Hiz is the duration of interruption of the i-th turbine generator in the hydropower station during the high water level period.

[0534] The high water level is the water level at which a hydropower station can generate electricity. If the contract stipulates that the hydropower station can generate electricity when the water level is in the range of 240-265m, then 240-265m is the high water level in this application.

[0535] Deep learning of water level data can reveal seasonal patterns in water level changes and facilitate further research. General rules can be derived from the characteristics of water level (or other quantities) data. Using the target formula or generalized electricity pricing scheme described in the previous examples, and based on fundamental data such as electricity costs and water consumption for power generation, correlated data, such as power plant revenue, can be obtained. Furthermore, features and correlations among water level, electricity costs, power plant revenue, power plant costs, and water consumption for power generation can be extracted to establish a model of the median water level and electricity price. Through deep learning and correlation analysis, a reasonable scheduling strategy for hydropower station revenue and water resource assets can be derived. This scheduling strategy is not limited to upstream quarterly power generation.

[0536] Table 12 is a data calculation table for the cascade hydropower stations provided in this application. The cascade hydropower stations in Table 12 include an upstream hydropower station and a downstream hydropower station, totaling two hydropower stations. The data in Table 12 are calculated based on the water resource data of each hydropower station in the cascade hydropower station at the current time, and formulas (2.1), (2.2), and (2.3).

[0537] Those skilled in the art can calculate the power generation revenue for other months in 2023 by referring to the electricity costs for different quarters and water levels provided in the foregoing embodiments, and by formulas (2.1), (2.2) and (2.3).

[0538] It should be noted that in Table 12, "upper dam" refers to the upstream reservoir; "lower dam" refers to the downstream reservoir; "upper powerhouse" refers to the powerhouse of the hydropower station corresponding to the upstream reservoir; and "lower powerhouse" refers to the powerhouse of the hydropower station corresponding to the downstream reservoir.

[0539] The water volume in Table 12 is in ten thousand cubic meters. 3 The unit for water level is meters (m); the unit for flow rate is cubic meters per second (m³). 3 / s; rainfall is measured in mm; electricity is measured in ten thousand kWh. The meanings of the letters in Table 6 are explained in the preceding embodiments and will not be repeated here.

[0540] Table 12

[0541] In one optional implementation, the target historical data for each hydropower station is obtained as follows:

[0542] First, calculate the historical power generation water consumption and historical power plant revenue for each month across N historical years. Here, N is an integer greater than or equal to 1.

[0543] For example, the inflow of water into hydropower station A in a specific month, such as January 2024, can be obtained; the reservoir capacity of hydropower station A in January 2024 can be obtained; the water consumption of hydropower station A in January 2024 can be obtained by subtracting the reservoir capacity from the inflow. The power generation and power generation duration of hydropower station A in January 2024 can be calculated; then, the water consumption of hydropower station A in January 2024 can be divided by the power generation of hydropower station A in January 2024, and then divided by the power generation duration of hydropower station A in January 2024, to obtain the historical power generation and water consumption of hydropower station A in January 2024.

[0544] Figure 6B is a schematic diagram of the historical water consumption for power generation of an upstream hydropower station for each month, provided in an embodiment of this application. In Figure 6B, the horizontal axis represents cubic meters per kilowatt-hour; the vertical axis represents months. Figure 6B shows the historical water consumption for power generation of the upstream hydropower station from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical water consumption figures calculated based on historical data.

[0545] Figure 6C is a schematic diagram of the historical water consumption for power generation of a downstream hydropower station for each month, provided in an embodiment of this application. In Figure 6C, the horizontal axis represents cubic meters per kilowatt-hour; the vertical axis represents months. Figure 6C shows the historical water consumption for power generation of the downstream hydropower station from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical water consumption figures calculated based on historical data.

[0546] Based on the comparison of power generation water consumption in Figures 6B and 6C, it can be clearly seen that quarterly regulation is more suitable for upstream areas, while day-ahead regulation (also known as day-ahead dispatch) is more suitable for downstream areas.

[0547] As described in the previous embodiments, the sum of formulas (2.1) and (2.3) represents the actual power plant revenue of the hydropower station. Therefore, by substituting historical data into formulas (2.1) and (2.3), the historical power plant revenue can be obtained.

[0548] Figure 6D is a schematic diagram of the historical power plant revenue for each month of an upstream hydropower station according to an embodiment of this application. The horizontal axis in Figure 6D is in US dollars; the vertical axis is for months. Figure 6D shows the historical power plant revenue for each month of the upstream hydropower station from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical power plant revenues calculated based on historical data.

[0549] Figure 6E is a schematic diagram of the historical power plant revenue for each month of a downstream hydropower station according to an embodiment of this application. The horizontal axis in Figure 6E is in US dollars; the vertical axis is for months. Figure 6E shows the historical power plant revenue for each month of the downstream hydropower station from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical power plant revenues calculated based on historical data.

[0550] Furthermore, water asset data for each power plant can be calculated for each month. The water assets of each power plant in this application refer to the sum of formulas (2.1), (2.2), and (2.3).

[0551] Figure 6F is a schematic diagram of the water assets of an upstream hydropower station for each month, provided in an embodiment of this application. The horizontal axis in Figure 6F represents the quantity of water assets in US dollars; the vertical axis represents the units of months. Figure 6F shows the historical water assets of the upstream hydropower station for each month from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical water assets extrapolated from historical data.

[0552] Figure 6G is a schematic diagram of the water assets of a downstream hydropower station for each month according to an embodiment of this application. In Figure 6G, the horizontal axis represents the quantity of water assets in US dollars; the vertical axis represents the units of months. Figure 6G shows the historical water assets of the downstream hydropower station for each month from January to December 2023 and from January to December 2024. The data for November and December 2024 are historical water assets extrapolated from historical data.

[0553] For ease of understanding, Figures 6B, 6C, 6D, 6E, 6F, and 6G of this application illustrate the historical power generation water consumption, historical power plant revenue, and historical water assets (revenue) for each hydropower station in a cascade hydropower system comprising an upstream and a downstream hydropower station, for each month. Those skilled in the art can obtain the historical power generation water consumption, historical power plant revenue, and historical water assets (revenue) for each hydropower station in other cascade hydropower systems by referring to the methods described in the foregoing embodiments.

[0554] Then, for each month, a map is plotted based on the N historical power generation and water consumption data corresponding to that month to obtain the historical power generation and water consumption map for that month.

[0555] For example, for January, we can obtain the historical power generation water consumption for January 2010, January 2011, January 2012, and even January 2024; and generate a graph with the year on the horizontal axis and the historical power generation water consumption on the vertical axis as the historical power generation water consumption map for January.

[0556] Similarly, for February, we can obtain the historical power generation and water consumption for February 2010, February 2011, February 2012, and even February 2024; and generate a graph with the year on the horizontal axis and the historical power generation and water consumption on the vertical axis, which serves as the corresponding historical power generation and water consumption map for February.

[0557] For each month, based on the historical power plant revenue of N corresponding months, a historical power plant revenue map for that month is obtained.

[0558] For example, for January, we can obtain the historical power plant revenue for January 2010, January 2011, January 2012, and even January 2024; and generate a graph with the year on the horizontal axis and the historical power plant revenue on the vertical axis as the historical power plant revenue map for January.

[0559] Among them, the historical power generation water consumption map represents the relationship between historical power generation water consumption and time; the historical power plant revenue map represents the relationship between historical power plant revenue and time.

[0560] Finally, the average power generation and water consumption corresponding to the historical power generation and water consumption map of the current month, and the average power plant revenue corresponding to the historical power plant revenue map of the current month, are used as target historical data.

[0561] The average power generation water consumption is the average of the historical power generation water consumption included in the historical power generation water consumption map. The average power plant revenue is the average of the historical power plant revenue included in the historical power plant revenue map.

[0562] After obtaining the current actual data and the target historical data, a determination is made based on the difference between the current actual data and the target historical data to determine whether to carry out water resource scheduling. Specifically:

[0563] Determine whether the difference between the actual power generation water consumption and the average power generation water consumption is within the preset allowable value for the difference in power generation water consumption (first judgment condition), and determine whether the difference between the actual power plant revenue and the average power plant revenue is within the preset allowable value for the difference in power plant revenue (second judgment condition).

[0564] If both the first and second judgment conditions are met simultaneously, water resource scheduling is not required, and the process returns directly to S601; if neither of the above two judgment conditions is met simultaneously, water resource scheduling is considered necessary, and the process proceeds to S604.

[0565] Furthermore, the actual water assets of the hydropower station in the current month can be calculated (the sum of formulas (2.1), (2.2), and (2.3)); the average water assets of the hydropower station in the current month can be obtained; it can be determined whether the difference between the actual water assets and the average water assets in the current month is within the preset allowable value for water asset difference (third judgment condition), and whether water resource scheduling is required.

[0566] That is, if the first, second, and third judgment conditions are met, it is determined that no water resource scheduling is required, and the process returns directly to S601; otherwise, it is considered that water resource scheduling is required, and the process proceeds to S604.

[0567] It should be noted that this application does not limit the preset allowable values ​​for power generation water use differences, preset allowable values ​​for power plant revenue differences, and preset allowable values ​​for water asset differences. Those skilled in the art can limit the above values ​​as needed.

[0568] S604 If it is determined that water resource scheduling will be carried out, a water resource scheduling scheme shall be determined based on the water resource data and water resource scheduling model of each hydropower station.

[0569] Among them, the water resource scheduling model is a model that is trained in advance and used to output water resource scheduling schemes.

[0570] In one alternative implementation, the training steps of the water resource scheduling model include:

[0571] The first step is to obtain training data.

[0572] The training data in this application includes sample water resource data obtained by each hydropower station in the cascade hydropower station at the sample time point, the electricity price scheme at the sample time point, and the sample water resource scheduling scheme at the sample time point.

[0573] It should be noted that the sample time point in this application is any one of multiple time points during the operation of the hydropower station under all operating conditions. In other words, the sample time point can be a time point when the hydropower station is operating normally, and the data corresponding to the sample time point is the data corresponding to the normal operating time of the hydropower station; the sample time point can also be a time point when the hydropower station conducts load shedding tests, water conservation tests, or vibration tests, and the data corresponding to the sample time point is the data corresponding to the hydropower station under the above-mentioned test conditions.

[0574] The sample water resources data includes historical power generation water consumption, historical power plant revenue, and historical average water assets; the sample water resources dispatching schemes include day-ahead dispatching and seasonal dispatching.

[0575] Understandably, water resource allocation for each hydropower station in a cascade hydropower system is a highly complex process. At different times, and given variations in factors such as the water level in the reservoir where each hydropower station is located, the actual operating parameters of the power generation facilities, and rainfall, different water resource allocation schemes will be employed.

[0576] It is understandable that the upstream hydropower station in a cascade hydropower station has a larger storage capacity than the downstream hydropower station; the water flow regulation space of the upstream hydropower station is also larger than that of the downstream hydropower station. Based on these reasons, in the water resource scheduling schemes included in the training data provided in this application, the upstream hydropower station adopts a quarterly regulation method (also known as quarterly scheduling and quarterly regulation), while the downstream hydropower station adopts a day-ahead regulation method (also known as day-ahead scheduling and day-ahead regulation).

[0577] Quarterly regulation refers to adjusting the reservoir capacity of a hydropower station to store a portion of the runoff during the wet season for use during the dry season. This regulation method enables the hydropower station to effectively redistribute water resources throughout the year, especially during seasons with uneven water distribution. Quarterly regulation allows for the storage of excess water during seasons of ample water availability and the release of stored water during seasons of scarcity, ensuring the stable operation and continuous power generation of the hydropower station.

[0578] Day-ahead scheduling refers to the redistribution of runoff within a 24-hour period. This regulation method is mainly used to address short-term (e.g., within a day) changes in water resources and fluctuations in electricity demand. Through day-ahead scheduling, hydropower stations can flexibly adjust their power generation plans and water resource allocation based on real-time water resource conditions and electricity demand to ensure the stability and reliability of power supply.

[0579] Quarterly scheduling focuses on the balanced distribution of water resources between seasons to achieve long-term stable operation and power generation efficiency; while day-ahead regulation focuses more on the dynamic balance of water resources and electricity demand in the short term to ensure the stability and reliability of power supply. Hydropower stations comprehensively adopt these two regulation methods based on their own reservoir capacity, water resource conditions, and electricity market demand to achieve efficient utilization of water resources and optimization of power supply.

[0580] The calculation of water assets is meaningful precisely because of the day-ahead scheduling downstream and the seasonal scheduling upstream. Both aim to achieve a balance between cost and economy. Downstream focuses on power generation, while upstream focuses on controlling the water volume and level of the entire basin. As shown in Figures 6B and 6C, in the comparison of day-ahead and quarterly scheduling between upstream and downstream, the water consumption per kilowatt-hour is 0.018 m³. 3 / kWh and 0.026m 3 / kWh, because the focus is currently on monthly profitability, it indicates that downstream power generation efficiency is better, while upstream quarterly scheduling focuses more on the overall situation of the entire basin, taking power generation into account. Furthermore, comparing the monthly cumulative water assets of upstream and downstream, it is clear that the details of water assets upstream are more numerous than those upstream, further illustrating the importance of the quarterly conditions adopted upstream.

[0581] By abstracting and analyzing the high-order features of the data, especially the research on power generation water consumption and water assets, the two scheduling methods can implement different scheduling strategies for the upstream and downstream of the hydropower station under the premise of ensuring the safety of the smart hydropower station, that is, under the condition that all operating conditions meet the requirements of load shedding, variable power vibration measurement and water conservation safety.

[0582] Figure 6H is a schematic diagram of a water resource regulation scheme for a cascade hydropower station provided in an embodiment of this application. The cascade hydropower station includes an upstream hydropower station and a downstream hydropower station. The letters in Figure 6H have the following meanings: J1 represents the cost price of electricity; S1 represents the measured water level of the upstream hydropower station; S2 represents the measured water level of the downstream hydropower station; R represents the minimum water level at which the downstream hydropower station is allowed to generate electricity; Q represents the minimum water level at which the upstream hydropower station is allowed to generate electricity; J represents the current grid connection price, which is also the basic value of the preset grid connection price mentioned in the aforementioned embodiment.

[0583] Referring to Figure 6H, the specific water resource regulation scheme for this cascade hydropower station is as follows:

[0584] S401, determine the cost electricity price J1, the measured water level of the upstream hydropower station S1, and the measured water level of the downstream hydropower station S2.

[0585] S402, determine whether S2 < R.

[0586] Determine whether the measured water level of the downstream hydropower station is lower than the minimum water level at which the downstream hydropower station is allowed to generate electricity. If S2 is less than R, proceed to S403; otherwise, proceed to S406.

[0587] S403, determine whether S1 > Q and cost electricity price < grid connection electricity price.

[0588] Specifically, it determines whether the measured water level at the upstream hydropower station is higher than the minimum water level allowed for power generation at the upstream hydropower station, and whether the cost electricity price is lower than the grid connection price (the grid connection transaction price at the current moment). If yes, proceed to S404; otherwise, proceed to S407.

[0589] S404 is used for quarterly regulation of upstream hydropower stations.

[0590] S405 is used for day-ahead regulation of downstream hydropower stations.

[0591] S406, awaiting shutdown instructions from the power grid in case of a fault or water shortage at the power plant.

[0592] S407, determine whether the grid-connected electricity price changes J = J ± 0.001, and the cost price J1 < J.

[0593] If yes, proceed to step S404; otherwise, proceed to step S408.

[0594] S408, high-frequency water level regulation, variable water level.

[0595] It should be emphasized that the water resource regulation scheme shown in Figure 6H is only for ease of understanding and represents a possible water resource scheduling scheme in actual operation. Different targeted water resource regulation schemes will be adopted for different cascade hydropower stations, at different times, and when the water resource data of each hydropower station is different.

[0596] The second step is to input the training data into the deep learning model to obtain the predicted water resource adjustment scheme corresponding to the training data.

[0597] For example, convolutional neural networks, long short-term memory networks, and gated recurrent units can be used as deep learning models in this application.

[0598] The sample water resource data obtained by each hydropower station in the cascade hydropower stations in the training data at the sample time point, and the electricity price scheme at the sample time point are input into the deep learning model. The deep learning model then inputs the predicted water resource adjustment scheme.

[0599] The third step involves using the sample water resource scheduling scheme as the true value, determining the backpropagation error of the sample water resource deep adjustment scheme and the predicted water resource adjustment scheme through the loss function, and updating the network parameters of the deep learning model based on the backpropagation error until the backpropagation error reaches the set value, thus obtaining the trained water resource scheduling model.

[0600] In this application, the sample water resource scheduling schemes at sample time points in the training data are taken as ground truth. The difference between the sample water resource deep adjustment scheme and the predicted water resource adjustment scheme, i.e., the backpropagation error, is calculated through a loss function. The backpropagation error is used as the loss value, and the network parameters in the deep learning model, such as weights and biases, are updated through the loss value to reduce future backpropagation errors. Training stops when the latest backpropagation error reaches a set value, thus obtaining a trained water resource adjustment model.

[0601] Loss functions are fundamental to guiding effective model learning. Different loss functions can be selected or designed based on different tasks, enabling the model to extract valuable information from the data. Below, we introduce several common loss functions, including: squared loss, cross-entropy loss, mean squared error, and mean absolute error. This application does not limit the specific form of the loss function. Loss functions and optimization algorithms are two important components of machine learning. Based on the loss function, the selected optimization algorithm can also be chosen according to actual needs. For example, gradient descent can be used, which utilizes gradient information and iteratively adjusts parameters to find a suitable solution.

[0602] In summary, this application presents a water resource scheduling method for cascade hydropower stations, comprising: acquiring water resource data of each hydropower station in the cascade hydropower station at the current moment; for each hydropower station, determining whether the hydropower station is in a safe operating state based on its water resource data; when it is determined that all hydropower stations are in a safe operating state, determining whether to carry out water resource scheduling based on the water resource data of each hydropower station, a preset electricity price scheme, and target historical data; if it is determined that water resource scheduling should be carried out, then determining a water resource scheduling scheme based on the water resource data of each hydropower station and a water resource scheduling model. In this way, by replacing manual experience with a water resource scheduling model, the optimal water resource scheduling scheme can be determined based on the water resource data of each hydropower station at different times, while ensuring the safe operation of the hydropower station, thereby maximizing power generation revenue.

[0603] It should be noted that, in addition to the method for determining whether each hydropower station in the cascade hydropower station is in a safe operating state mentioned in the foregoing embodiments of this application, it is also necessary to conduct load shedding tests on each reservoir in the cascade hydropower station before the cascade hydropower station is officially put into operation to determine whether the operating parameters of each hydropower station in the cascade hydropower station meet the preset load shedding test index requirements under emergency conditions; that is, to determine whether the power generation equipment of each hydropower station in the cascade hydropower station can work normally under emergency conditions, thereby verifying the dynamic performance of the equipment under extreme conditions.

[0604] Load shedding occurs when the electricity load of end users decreases (e.g., due to a failure of large electrical equipment or a large-area power outage caused by a line fault), and the power generation of the hydro generator at the power plant exceeds the amount transmitted to users. In this case, the power plant is required to reduce the power generation to a value that is appropriate for the actual load. Alternatively, it may be due to internal reasons at the power plant, such as a sudden tripping of the circuit breaker at the power grid outlet, causing the load of the hydro generator to suddenly drop to almost zero. These are the actions taken by the power plant.

[0605] After a hydro-generator unit sheds load, the enormous residual energy causes the unit speed to rise rapidly. The governor quickly closes the guide vanes, and after a period of adjustment, it regains stable operation under no-load conditions. During the load shedding process, in addition to the maximum speed increase and the maximum water hammer pressure increase, it is also necessary to assess the quality of the dynamic process indicators of load shedding.

[0606] The load shedding test parameters in this application include, but are not limited to: speed rise rate, speed adjustment time, speed overshoot, volute water pressure rise rate, guide angle closing time, tailrace tube vacuum degree, etc.

[0607] It is understandable that when a cascade hydropower station conducts a load shedding test, the cascade hydropower station is considered to have passed the load shedding test only if each operating parameter during the test, such as the rate of increase in speed, the speed regulation time, the speed overshoot, the rate of increase in water pressure in the spiral casing, the guide angle closing time, and the tailrace vacuum degree, meets the corresponding load shedding test indicators. Otherwise, the cascade hydropower station needs to be overhauled until it can meet the requirements of the load shedding indicators before the contents of S601-S604 in this application are implemented.

[0608] It is important to emphasize that the water conservation tests, vibration tests, and load shedding tests mentioned in the preceding embodiments can obtain a large amount of experimental data. This data can be digitized and become a data resource. After obtaining this large amount of experimental data, i.e., a large data resource, various required deep learning models can be trained based on deep learning. After obtaining the current water resource data, the trained deep learning model can generate various predictive indicators indicating the safety status of the hydropower station, including but not limited to: speed regulation time, guide angle closing time, etc. Comparing these predictive indicators with the actual indicators specified in the contract allows for efficient determination of whether there are any safety hazards at the hydropower station and provides a relatively accurate prediction of the hydropower station's safety status as indicated by the current water resource data. This effectively avoids various risks that may arise during the power generation process of the hydropower station, ensuring that the turbines and generators of each hydropower station in the cascade hydropower station operate under safe conditions.

[0609] Figure 6I is a schematic diagram of the interaction between a power plant and the power grid according to an embodiment of this application. Figure 6I shows the interaction between a cascade hydropower station and the power grid, including an upstream hydropower station and a downstream hydropower station. The power plant side in Figure 6I includes the upstream hydropower station and the downstream hydropower station. Among them, the upstream hydropower station performs seasonal scheduling, and the downstream hydropower station performs day-ahead scheduling.

[0610] As shown in Figure 6I, the upstream hydropower station analyzes the characteristic production factor data to identify the seasonal variation pattern of water level and determine the median water level for each season. Based on the determined median water level, the upstream hydropower station calculates the electricity price operator function (i.e., PR). 实 On the one hand, the upstream hydropower station notifies the downstream hydropower station to determine the median water level of the downstream hydropower station. The upstream hydropower station calculates the power plant revenue (power generation revenue) and water consumption per kilowatt-hour (water consumption per kilowatt-hour) based on the water level and electricity price operator function, and sends the calculated power plant revenue and water consumption per kilowatt-hour to the downstream hydropower station.

[0611] Downstream hydropower stations determine the median water level and the electricity price operator function (i.e., PR). 实 At the same time, it receives the power plant revenue and water consumption per kilowatt-hour from the upstream hydropower station, calculates the total power plant revenue and total water consumption per kilowatt-hour of the cascade hydropower stations, and sends the calculated data to the power grid side.

[0612] It should be noted that the power grid needs to calculate grid stability in real time. Only when the grid is stable can subsequent work be carried out, including receiving the total power plant revenue and total electricity and water consumption from downstream hydropower stations, performing deep learning and updating based on the scheduling strategy library, and feeding the scheduling strategy back to the downstream hydropower stations.

[0613] Next, downstream hydropower stations sequentially perform safety operation calculations for their downstream units (i.e., calculate the safety stability of downstream units), execute cleared power output and unit actions, and send relevant information to the grid side. Upstream hydropower stations sequentially perform safety stability calculations for their upstream units (i.e., calculate the safety stability of upstream units), send the calculation results to the grid side, and apply for traded power volume and price. The grid side receives and records this data. Based on the results of the downstream cleared power output and the upstream applied power volume, it performs another grid stability calculation. Because upstream hydropower stations affect the safety of the entire waterway, the value of the upstream applied power volume is more important than the value of the downstream cleared power volume during the grid stability calculation based on the results of the downstream cleared power output and the upstream applied power volume.

[0614] After the power grid is confirmed to be stable, the power grid side sends the cleared electricity price and power dispatch instructions to the upstream hydropower station; the upstream hydropower station executes the cleared power and unit operation; the upstream and downstream hydropower stations calculate hydropower assets respectively.

[0615] Figure 6J is a flowchart of another water resource scheduling method provided in an embodiment of this application. Referring to Figure 6J, the other water resource scheduling method of this application includes:

[0616] S1601, Obtain hydrological data from upstream and downstream reservoirs as basic data.

[0617] S1602 performs digital calculations of upstream and downstream reservoir capacity and flow.

[0618] S1603 performs digital calculations for water conservation safety.

[0619] S1604, Perform safety calculations for hydropower generation.

[0620] S1605, based on upstream and downstream hydrological data, determine whether water resource allocation is necessary.

[0621] If water resource allocation is required, proceed to S1606; otherwise, return to S1602.

[0622] S1606 proposes a scheduling scheme based on upstream and downstream hydrological data and a deep learning model.

[0623] The scheduling scheme is the water resource scheduling scheme; the downstream hydropower station implements day-ahead scheduling, the upstream hydropower station implements quarterly scheduling (also known as seasonal scheduling), and indicates whether to carry out high-frequency scheduling of water level in the hydropower station.

[0624] The contents of S1601-S1606 are described in S601-S604 and S401-S408, and will not be repeated here.

[0625] Embodiment 4 in this application aptly illustrates the application of the concepts of symbiosis, migration, and emergence in the digital power system constructed by the technical solution of this application. When selecting quarterly dispatch, a large amount of data is first acquired, including water levels. Then, the patterns in the data are studied, and finally, the data is generalized into a process or electricity price operator function based on the median water level. This function is then packaged into an intelligent agent, realizing the specific function of a multifunctional intelligent agent in the marketization of quarterly dispatch in hydropower. Similarly, other data features can be discovered and abstracted, so that the operators in the intelligent agent each possess their own specific functions.

[0626] In this embodiment, based on the relationship between water level and electricity price, a relationship between natural production factors and power grid production control factors is established, so that the electricity price changes with the water level. At the same time, water level scheduling and control, such as seasonal scheduling, are data-driven, so that both reservoir safety and economy can be considered. Meanwhile, water level control and power generation are directly related to data. The seasonal characteristics abstracted from water level characteristics, i.e., the median water level, are solidified with electricity price, so that market-based power generation can change with natural production factors, which is in line with natural and economic laws.

[0627] In this embodiment, all operating conditions are demonstrated through experiments and operation, and stored as data samples for continuous comparison. This is a data training method based on data assets, with water resources serving as the data asset. The power generation cost is determined by historical data, comparing the corresponding hydroelectric power costs at different water levels. Safety is determined by historical operation and experiments; the experimental power data is used as samples to characterize all operating conditions, ensuring the safety of conditions falling within the samples. Under regular equipment inspections, equipment safety is guaranteed. Furthermore, system safety is ensured through tests such as water conservation, variable power vibration measurement, and load shedding.

[0628] Example 5

[0629] In Example 5, the power function requirement for realizing grid-side dispatch control over power plants specifically refers to the grid dispatch of power plants in a large-scale, small-scale power system. The functional operator is the large-scale, small-scale dispatch operator, which is used to realize the grid dispatch of power plants in the large-scale, small-scale power system. Example 5 describes the process of using an intelligent agent built based on the large-scale, small-scale dispatch operator in a digital power system.

[0630] The current structure of large-scale generator-small-grid power systems may lead to stability issues in the face of large disturbances, especially frequency stability. The grid's ability to withstand large disturbances is weak. Regional load fluctuations, unit failures, or voltage collapses, especially in the event of large unit tripping or sudden load changes, can cause significant fluctuations in grid frequency and voltage, potentially even leading to grid collapse. In some cases, it can also cause regional voltage collapses, jeopardizing system security, causing system oscillations, and endangering the safety of large generating units. Currently, there is a lack of dispatching schemes to ensure the safety and stability of power systems with large-scale generator-small-grid structures.

[0631] To address the aforementioned issues, the inventors proposed a grid dispatching method for large-scale generator-small-grid power systems, where a digital power system based on a multi-functional intelligent agent invokes a large-scale generator-small-grid dispatching operator. In large-scale generator-small-grid scenarios, the impact of power generation points on the grid is even more significant. Different types of power generation points have different characteristics; therefore, voltage and frequency multi-round adjustment schemes were specifically designed. For power systems with large-scale generator-small-grid characteristics, grid dispatching utilizes three adjustment "tools": voltage multi-round adjustment, frequency multi-round adjustment, and power adjustment. Combined with pre-configured emergency reserve capacity, capacity support is provided for voltage and frequency adjustments under large-scale generator-small-grid conditions. Furthermore, the dynamic frequency characteristics of the entire grid are used as a stability target. Based on this, combined with the projected grid load and the characteristics of each power generation point, the unit output curves are generated, and the future output power of the units is dispatched. Through multi-round adjustments of voltage, frequency, and power, the safety and stability of the power system under large-scale generator-small-grid conditions are effectively guaranteed. By applying the technical solution of this application, the power supply stability and reliability of countries with power systems characterized by large generators and small grids can be effectively improved, and the grid collapse caused by events such as large generator trips or sudden changes in load can be reduced.

[0632] Figure 7A is a schematic diagram illustrating an application scenario of a power grid dispatching method in a large-scale, small-scale power system provided in an embodiment of this application. As shown in Figure 7A, in practical applications, various power sources of different generation types (e.g., photovoltaic, thermal power, wind power, nuclear power, hydropower, etc.) may communicate with the power grid and transmit power. These different types of power sources can be collectively referred to as the power plant side. In this embodiment, the power plant side can report its own data to the power grid side and provide timely feedback; the power grid side can perform dispatching on the power plant side, such as dispatching voltage, frequency, power, etc. As can be seen from Figure 7A, in the actual power production process, the power grid side and the power plant side need to conduct close dispatching to achieve better and more stable power supply and maintain the cost requirements of both parties.

[0633] In practical scenarios, the configuration of power plants is not limited to the types of power sources shown in Figure 7A. In other words, Figure 7A does not limit the types of power sources in the grid dispatching method for the large-scale generator-small-grid power system described in this application. Furthermore, due to differences in national power development levels, basic po...

Claims

The method for constructing a digital power system based on a multifunctional intelligent agent is characterized by: include: Extract characteristic production factor data corresponding to the characteristic production factor group from the power data on the power plant side; In addition, extract control production factor data corresponding to the control production factor group from the power grid side's power data; The characteristic production factor group includes characteristic production factors at multiple levels; the control production factor group includes control production factors from multiple aspects. Based on the power function requirements of the digital power system, a target data vector is constructed using the extracted characteristic production factor data, and key control production factor data associated with the power function requirements are determined from the extracted control production factor data. Based on the target data vector and the key control production factor data, an artificial intelligence algorithm is used to learn the mapping function between the target data vector and the key control production factor data, a functional operator with the mapping function as its core is constructed, and an intelligent agent is constructed based on the functional operator. Various intelligent agents corresponding to different power function requirements are added to the digital power system, and each intelligent agent serves as a power function implementation unit in the digital power system. The method according to claim 1, characterized in that, The characteristic production factor group includes primary, secondary, and tertiary characteristic production factors corresponding to various different power generation types. The primary characteristic production factors are natural characteristic production factors, the secondary characteristic production factors are single-system characteristic production factors, and the tertiary characteristic production factors are plant system characteristic production factors. The natural characteristic production factors directly reflect natural characteristics; the single-system characteristic production factors are production factors involving individual systems within the power system; and the plant system characteristic production factors are production factors involving the entire power system. The method according to claim 1, characterized in that, The various factors of production control mentioned above include: electricity price, inertia and frequency, active and reactive power, carbon emissions from electricity, safety and stability, and grid connection and disconnection. The extraction of control production factor data corresponding to the control production factor group from the power grid side includes: From the power data on the grid side, extract electricity price data, inertia and frequency data, power data, carbon emission data, safety and stability index data, and grid connection and disconnection impact data. The method according to claim 1, characterized in that, The method further includes: Configure the intelligent agent with a power plant-side interface, a grid-side interface, and an intelligent agent interface; The intelligent agent communicates with the power plant through the power plant-side interface; The intelligent agent communicates with the power grid through the power grid-side interface; The agent communicates with other agents through the agent interface. The method according to any one of claims 1-4, characterized in that, The power function requirements are as follows: The power function requirements for power transactions between the grid and power plants, the power function requirements for grid-side dispatch and control of power plants, or the power function requirements for grid self-regulation. The method according to claim 5, characterized in that, The specific power function requirement for power trading between the grid side and the power plant side is to calculate electricity prices. The functional operator is an electricity price operator, used to calculate electricity prices. The process by which the digital power system utilizes an intelligent agent built based on the electricity price operator includes: The intelligent agent is invoked to obtain the electricity price calculation result on the power plant side through the electricity price operator; The electricity price calculation results are verified using a financial verification model on the grid side. The financial verification model includes: a first verification condition and a second verification condition. The first verification condition is: the actual output of various types of power plants is greater than or equal to the grid power volume. The second verification condition is: the sum of the products of the actual output of various types of power plants, the grid electricity price at the corresponding time, and the power generation duration is less than or equal to the product of the grid average price and the grid power volume. If the electricity price calculation result meets both the first verification condition and the second verification condition, then the electricity price calculation result is determined to have passed verification; if the electricity price calculation result does not meet either the first verification condition or the second verification condition, then the electricity price calculation result is determined to have failed verification. If the electricity price calculation result is verified, then a suitable and feasible electricity price calculation function for the grid side is selected from the scheduling strategy library, and the electricity price calculation function is loaded into the electricity price operator to update the electricity price operator. The updated electricity price operator is used to obtain the electricity price calculation results on the grid side. The method according to claim 6, characterized in that, After loading the electricity price calculation function into the electricity price operator to update the electricity price operator, the process of using the intelligent agent constructed based on the electricity price operator further includes: The grid side records the applied transaction volume and price reported by the power plant to the grid side; Based on the grid-side determination of whether grid stability needs to be maintained, it is determined whether the virtual power plant function needs to be activated. If it is determined that the virtual power plant function does not need to be activated, the grid side sends dispatch instructions on the clearing price and power volume to each type of power plant based on the grid side's electricity price calculation results. If it is determined that the virtual power plant function needs to be activated, then the virtual power plant function is activated to utilize its storage capacity; The power plant receives the clearing price from the grid. The power plant cost and the clearing price are combined and verified using a financial verification model on the power plant side. The financial verification model on the power plant side includes a third verification condition. The third verification condition is: the sum of the products of the power plant's actual output, the grid price at the corresponding time, and the power generation duration is greater than the product of the power plant cost and the power plant's actual output. If the verification result indicates that the verification is successful, then further determine whether different power sources of the same power plant type can be cleared based on power volume and electricity price; If all power points can be cleared, then clear the power consumption. The method according to claim 5, characterized in that, The power function requirement for power trading between the grid side and the power plant side is specifically power clearing, and the functional operator is a clearing operator, which is used for power clearing; the process by which the digital power system utilizes an intelligent agent built based on the clearing operator includes: Based on the power generation element characteristics of each associated power plant in the power grid, low-level data abstraction processing is performed to obtain the first data vector of each power generation element characteristic. Long-term transaction electricity prices are predicted based on a preset long-term electricity price function library and each of the first data vectors, so as to establish a long-term transaction stack. A spot trading stack is established based on the spot trading electricity prices of each of the associated power plants, and a first clearing function is determined through the long-term trading stack and the spot trading stack. Based on the first clearing function, determine the power supply and demand balance state of the power grid when it is cleared under the first clearing function; The first clearing function is adjusted according to the power supply and demand balance state to obtain the second clearing function, and power clearing is performed based on the second clearing function. The method according to claim 8, characterized in that, The power supply and demand balance state includes: over-generation state; the first clearing function includes: first long-term trading volume; The step of adjusting the first clearing function according to the power supply and demand balance state to obtain the second clearing function includes: When the power supply and demand balance state is the over-generation state, the characteristics of each power generation element are subjected to high-order data abstraction processing by a preset deep learning model to obtain the second data vector of each associated power plant. The increase in long-term transaction volume is determined based on each of the second data vectors and the preset long-term electricity price function library; The first long-term trading volume is adjusted based on the increase in the long-term trading volume to obtain the second clearing function. The method according to claim 9, characterized in that, The power supply and demand balance state includes: under-generation state; the first clearing function includes: first spot trading volume; The step of adjusting the first clearing function according to the power supply and demand balance state to obtain the second clearing function includes: When the power supply and demand balance state is the under-generation state, for the first generation curve of the power grid when it is cleared based on the first clearing function, the first generation curve is fitted by a preset fitting function to obtain the fitted generation curve. Based on the fitted power generation curve and the preset planned power generation curve, determine whether the first spot trading volume meets the preset planned power generation curve; When the first spot trading volume does not meet the preset planned power generation curve, the preset long-term electricity price function library is modified by a preset reinforcement learning model to determine the reduction value of the long-term trading volume. The increase in spot trading volume is obtained by adjusting the proportion of spot trading volume in the first clearing function based on the decrease in long-term trading volume. The first spot trading volume is adjusted based on the increase in the spot trading volume to obtain the second clearing function. The method according to claim 10, characterized in that, When the first spot trading volume does not meet the preset planned power generation curve, the preset long-term electricity price function library is modified using a preset reinforcement learning model to determine the reduction value of the long-term trading volume, including: The preset long-term electricity price function library is screened using a preset reinforcement learning model to obtain the first long-term transaction electricity price function. The function index of the first long-term electricity price function is adjusted to determine the reduction value of the long-term transaction volume. The method according to claim 8, characterized in that, The step of predicting long-term transaction electricity prices based on a preset long-term electricity price function library and each of the first data vectors to establish a long-term transaction stack includes: Using the first data vector, the function indicators of the preset electricity price function library are adjusted in a data-driven manner to obtain a dynamic electricity price function library; Based on the dynamic electricity price function library and the characteristics of each of the power generation elements, long-term transaction electricity prices are predicted to obtain the long-term transaction predicted electricity prices of each of the associated power plants. The long-term transaction stack is established by predicting electricity prices through the long-term transactions of each of the associated power plants. The method according to claim 8, characterized in that, The establishment of a spot trading stack based on the spot trading electricity prices of each of the associated power plants includes: Based on the first data vector of each associated power plant, the predicted power of each associated power plant is obtained; A preliminary spot trading stack is established based on the predicted power generation of each of the associated power plants; Based on the spot trading electricity price of the associated power plant, the initial spot trading stack is adjusted to obtain the spot trading stack. The method according to claim 8, characterized in that, The step of performing low-level data abstraction processing based on the power generation element characteristics of each of the associated power plants to obtain a first data vector of the power generation element characteristics includes: Based on the characteristics of each of the power generation elements, an energy consumption data model is established for each of the associated power plants; the energy consumption data model is used to evaluate the power generation performance of the associated power plants. The energy consumption data model is used to perform low-level data abstraction processing on the characteristics of each power generation element to obtain the first data vector of each power generation element characteristic. The method according to claim 5, characterized in that, The power function requirement for realizing grid-side dispatch control over power plants specifically refers to peak-shaving dispatch in high-frequency load change scenarios for thermal power plants. The functional operator is a thermal power peak-shaving dispatch operator, which is used to realize peak-shaving dispatch in high-frequency load change scenarios for thermal power plants. The process by which the digital power system utilizes an intelligent agent constructed based on the thermal power peak-shaving dispatch operator includes: Calculate the heat storage margin of energy storage devices in a thermal power plant; the heat storage margin includes a heat storage margin characterization value and a heat release capacity characterization value; The system sends the correlation data between the load, performance, and cost of each thermal power unit in the thermal power plant, as well as the heat storage margin, to the grid side so that the grid side can determine the target units for expected auxiliary peak shaving and issue peak shaving dispatch instructions based on the changes in thermal power load demand, the correlation data provided by each thermal power plant, and the heat storage margin. The system receives a peak-shaving dispatch instruction from the power grid side; the peak-shaving dispatch instruction carries the unit identifier of the target unit and peak-shaving requirement information for the target unit; the peak-shaving requirement information includes the peak-shaving load curve. Assisted peak shaving services are performed based on the peak shaving requirement information. The method according to claim 15, characterized in that, The process of using an intelligent agent constructed based on the aforementioned thermal power peak-shaving scheduling operator also includes: Before determining the target units for the expected auxiliary peak shaving, the quotation information for providing the auxiliary peak shaving service to the thermal power units that meet the requirements of the corresponding peak load curves is calculated based on the peak load curves of each thermal power unit. The quotation information is sent to the power grid side; Based on changes in thermal power load demand, the associated data provided by each thermal power plant, and the available heat storage capacity, the power grid side determines the target units for expected auxiliary peak shaving, specifically as follows: Based on changes in thermal power load demand, the associated data provided by each thermal power plant, the heat storage capacity, and the pricing information, the grid side uses data-driven methods to weigh costs and indicators to determine the target units for expected auxiliary peak shaving. The method according to claim 16, characterized in that, The grid side, based on changes in thermal power load demand, the associated data provided by each thermal power plant, the heat storage capacity, and the pricing information, uses data-driven methods to weigh costs and indicators to determine the target generating units for expected auxiliary peak shaving, including: Based on the changes in thermal power load demand, the associated data provided by each thermal power plant, and the heat storage margin, the power grid side generates peak load curves for each thermal power unit and calculates the ramp-up coefficient for each thermal power unit. Based on the ramp coefficient of each thermal power unit, several thermal power units expected to assist in peak shaving have been preliminarily identified. If among the multiple thermal power units expected to assist in peak shaving, there is a thermal power unit whose price information is less than or equal to K times the threshold price, then the thermal power unit whose price information is less than or equal to K times the threshold price is determined as the target unit for expected auxiliary peak shaving; the value of K is in the range of 1.1 to 1.

3. The threshold price is a reference price obtained from the historical price information obtained through deep learning on the power grid side. The method according to claim 17, characterized in that, The process of using the intelligent agent constructed based on the thermal power peak-shaving scheduling operator further includes: after the preliminary determination of multiple thermal power units expected to assist peak shaving, the grid side sends a heat pre-schedule instruction to the multiple thermal power units expected to assist peak shaving. The target units among the multiple thermal power units expected to assist in peak shaving will perform heat pre-shaving based on the heat pre-shaving command, through the heat storage devices within their respective thermal power plants. The method according to claim 17, characterized in that, The peak shaving requirement information also includes ramp speed requirement information; the execution of auxiliary peak shaving services based on the peak shaving requirement information includes: Determine whether there are equipment or communication faults in the target unit, and determine whether the target unit meets the required ramp speed information; If there are no equipment or communication faults in the target unit, and the target unit meets the ramp-up speed requirements, then auxiliary peak shaving services are executed according to the peak load curve and ramp-up speed requirements in the peak shaving requirements information. If there is a device failure or communication failure in the target unit, a failure message is reported to the power grid side; If the target unit does not meet the ramp speed requirement, a notification message indicating that the operating condition is not met is reported to the power grid side. The power grid side adjusts the peak shaving requirement information in the peak shaving dispatch instruction based on the prompt information indicating that the operating conditions are not met. Based on the feedback from each thermal power plant regarding unmet operating conditions, the grid side relearns to raise the threshold price, thereby enabling more units to be targeted for auxiliary peak shaving. The method according to claim 17, characterized in that, The calculation method for the gradeability coefficient of thermal power units is as follows: The ramp rate, steam pressure, and rated power generation capacity were extracted from the correlation data between the load, performance, and cost of the thermal power unit. The climbing coefficient of the thermal power unit is calculated based on its climbing speed, steam pressure, rated power generation capacity, heat storage margin, and the required climbing time from the power plant side. The method according to claim 20, characterized in that, The heat storage margin is expressed as a percentage; the climb coefficient of the thermal power unit is calculated based on the unit's ramp coefficient, steam pressure, rated power generation capacity, heat storage margin, and the required ramp time from the power plant, including: Calculate the product of the thermal power unit's climbing speed, climbing time, steam pressure, and heat storage margin; Calculate the ratio of the product to the rated generating capacity of the thermal power unit, and use the ratio as the ramp-up coefficient of the thermal power unit. The method according to any one of claims 15-21 is characterized in that, The heat storage margin and the associated data are calculated based on the coal type parameters of the selected target coal type; the selection method for the target coal type includes: Based on the higher and lower heating values ​​of each of the multiple candidate coal types, obtain the percentage difference between the higher and lower heating values ​​of each candidate coal type. The coal type with the smallest percentage difference between high and low calorific values ​​among the multiple candidate coal types is identified as the target coal type. The method according to claim 22, characterized in that, The percentage difference between high and low calorific values ​​is a low-order feature obtained by the thermal power plant after extracting features based on the coal type parameters of the corresponding coal type. Based on changes in thermal power load demand, the associated data provided by each thermal power plant, and the available heat storage capacity, the grid side determines the target units for expected auxiliary peak shaving and issues peak shaving dispatch instructions, including: Based on the associated data provided by each thermal power plant and the heat storage margin, the power grid side extracts intermediate-level features, which include features reflecting the power plant's energy consumption level and heat energy reserve level. The power grid side integrates the coal price cost in the associated data and the intermediate-level features to obtain high-level features, which include the relationship between fuel, electricity, and electricity price. Based on the changes in thermal power load demand, the intermediate-order characteristics, and the high-order characteristics, the power grid side determines the target generating units for expected auxiliary peak shaving and issues peak shaving dispatch instructions. The method according to claim 5, characterized in that, The power function requirement for realizing grid-side dispatch control over power plant side specifically refers to water resource dispatching of cascade hydropower stations. The functional operator is a water resource dispatching operator, which is used to realize water resource dispatching of cascade hydropower stations. The process by which the digital power system utilizes an intelligent agent constructed based on the water resource dispatching operator includes: Obtain water resource data for each hydropower station in the cascade hydropower station system at the current moment; For each of the aforementioned hydropower stations, based on the water resource data of that hydropower station, it is determined whether the hydropower station is in a safe operating state; When it is determined that all the hydropower stations are in the safe operating state, it is determined whether to carry out water resource scheduling based on the water resource data of each hydropower station, the preset electricity price scheme and the target historical data; the target historical data includes the average power generation water consumption and the average power plant revenue corresponding to the month in which the current moment is located. If water resource scheduling is determined, a water resource scheduling scheme is determined based on the water resource data and water resource scheduling model of each hydropower station; the water resource scheduling model is a pre-trained model used to output the water resource scheduling scheme; the water resource scheduling scheme includes day-ahead scheduling and seasonal scheduling. The method according to claim 24, characterized in that, The safe operating status includes the water conservation safety status; For each of the hydropower stations, determining whether the hydropower station is in a safe operating state based on its water resource data includes: For each of the hydropower stations, the measured rate of velocity rise and the measured turbine hydrostatic pressure are obtained from the water resource data of that hydropower station. For each of the hydropower stations, if the measured rate of increase of velocity is less than the theoretical rate of increase of velocity, and the measured turbine hydrostatic pressure is less than the theoretical turbine hydrostatic pressure, then the hydropower station is determined to be in the water conservation safety state. If the measured rate of increase of velocity is greater than or equal to the theoretical rate of increase of velocity, or the measured turbine hydrostatic pressure is greater than or equal to the theoretical turbine hydrostatic pressure, then the hydropower station is determined not to be in the water conservation safety state. The method according to claim 24, characterized in that, The safe operating state includes a vibration safe state. For each hydropower station, determining whether it is in a safe operating state based on its water resource data includes: For each of the aforementioned hydropower stations, the power of the turbine engine is obtained from the water resource data of that hydropower station and denoted as the first power. Determine whether the first power is within a preset power range; If the first power is within the preset power range, then the first power is adjusted to the second power; the second power is outside the preset power range. If the first power is outside the preset power range, then it is determined whether the hydropower station is in the vibration safety state. The method according to claim 26, characterized in that, The determination of whether the hydropower station is in the vibration safety state includes: For each of the hydropower stations, the actual values ​​of the turbine engine speed and vibration of different parts of the turbine engine are obtained from the water resource data of the hydropower station. For each of the hydropower stations, if the actual vibration value of each part is less than the permissible vibration value corresponding to that part at the specified rotational speed, then the hydropower station is determined to be in the vibration safety state. Otherwise, it is determined that the hydropower station is not in the aforementioned vibration safety state. The method according to claim 24, characterized in that, Before acquiring the water resource data of each hydropower station in the cascade hydropower stations at the current moment, the process of using the intelligent agent constructed based on the water resource scheduling operator further includes: A load shedding test is conducted on the cascade hydropower stations to determine whether the operating parameters of each hydropower station in the cascade hydropower stations meet the requirements of the preset load shedding test indicators. The method according to claim 24, characterized in that, The step of determining whether to conduct water resource scheduling based on the water resource data of each hydropower station, the preset electricity pricing scheme, and the target historical data includes: Based on the water resource data of each hydropower station and the preset electricity price scheme, the actual power generation water consumption and the actual power plant revenue of each hydropower station are determined as the current actual data; Based on the difference between the current actual data and the target historical data, a determination is made as to whether to carry out water resource scheduling. The method according to claim 29, characterized in that, The determination of the actual water consumption for power generation and the actual power plant revenue of each hydropower station, based on the water resource data of each hydropower station and the preset electricity pricing scheme, includes: For each hydropower station, the actual power generation and actual water consumption for a target time period are obtained from the water resource data of that hydropower station; the target time period is a past period ending at the current time. For each of the hydropower stations, the actual power generation and actual water consumption during the target time period are calculated. For each of the hydropower stations, the actual power plant revenue for the target time period is determined based on the electricity pricing scheme corresponding to the target time period and the actual water level of the hydropower station. The method according to any one of claims 24-30 is characterized in that, The method for obtaining the target historical data includes: Calculate the historical power generation and water consumption and historical power plant revenue for each month over N historical years; where N is an integer greater than or equal to 1. For each month, based on N historical power generation water consumption data for that month, a historical power generation water consumption map for that month is obtained; and based on N historical power plant revenue data for that month, a historical power plant revenue map for that month is obtained; the historical power generation water consumption map represents the relationship between historical power generation water consumption and time; the historical power plant revenue map represents the relationship between historical power plant revenue and time. The average power generation and water consumption corresponding to the historical power generation and water consumption map of the current month, and the average power plant revenue corresponding to the historical power plant revenue map of the current month, are used as the target historical data. The method according to any one of claims 24-30 is characterized in that, The training steps of the water resource allocation model include: Acquire training data; the training data includes sample water resource data of each hydropower station in the cascade hydropower stations at a sample time point, the electricity pricing scheme at the sample time point, and the sample water resource scheduling scheme at the sample time point; the sample time point is any one of multiple time points in the operation of the hydropower station under all operating conditions; the sample water resource data includes historical power generation water consumption, historical power plant revenue, and historical average water assets; the sample water resource scheduling scheme includes day-ahead scheduling and seasonal scheduling; The training data is input into a deep learning model to obtain a predicted water resource adjustment scheme corresponding to the training data; Using the sample water resource scheduling scheme as the true value, the backpropagation error of the sample water resource deep adjustment scheme and the predicted water resource adjustment scheme is determined by the loss function, and the network parameters of the deep learning model are updated based on the backpropagation error until the backpropagation error reaches the set value, thus obtaining the water resource scheduling model. The method according to claim 5, characterized in that, The power function requirement for realizing grid-side dispatch control over power plants specifically refers to grid-side dispatch of power plants in a large-scale power system with small-scale grids. The functional operator is a large-scale grid dispatch operator, which is used to realize grid-side dispatch of power plants in the large-scale power system. The process of the digital power system utilizing an intelligent agent built based on the large-scale grid dispatch operator includes: Based on the digital feature extraction of the power grid and power sources of various power generation types in the power system, the emergency reserve capacity of various power generation types that should be reserved in the power system is determined and the emergency reserve capacity is configured. Based on the output of the power sources of the various power generation types mentioned above, determine whether the current power system conforms to the characteristics of large generators and small grids; If it is determined that the current power system conforms to the characteristics of a large generator and a small grid, then the power grid utilizes the configured emergency reserve capacity for various power generation types to schedule power sources within the power system for voltage adjustment based on a multi-round voltage adjustment scheme, and / or schedules power sources within the power system for frequency adjustment based on a multi-round frequency adjustment scheme. The multi-round voltage adjustment scheme includes: voltage adjustment methods, voltage adjustment ranges, and voltage adjustment priority information for the various power generation types; the multi-round frequency adjustment scheme includes: frequency adjustment methods, frequency adjustment ranges, and frequency adjustment priority information for the various power generation types. Based on the stability control requirements of the frequency dynamic characteristic index of the entire power system, the expected load of the entire network and the characteristics of the generating units at each power source, the power grid generates the output curves of the generating units at each power source in the future period; the frequency dynamic characteristic index is used to numerically characterize the amount of power change of the entire network required to cause a unit frequency change in the power system. The power grid sends generation dispatch instructions to each power source to adjust power output; the generation dispatch instructions include the output curves of the corresponding power source units. The method according to claim 33 is characterized in that, The voltage multi-round adjustment range includes: photovoltaic, hydropower, nuclear power, thermal power, and energy storage; the voltage multi-round adjustment priority information includes: photovoltaic > hydropower > nuclear power > thermal power > energy storage; The multi-round voltage adjustment includes: voltage adjustment of the four basic rounds corresponding to photovoltaic, hydropower, nuclear power and thermal power, and voltage adjustment of the one-round accident round corresponding to fault conditions; The multi-round voltage adjustment method includes: the operating voltage and delay of the first basic round, the operating voltage level and delay of the remaining basic rounds, and the operating voltage and delay of the fault round. The method according to claim 34, characterized in that, The operating voltage of the first basic wheel is 0.84 pu, with a delay of 15-20 seconds; The voltage level difference for the remaining basic wheels is 0.01 pu, with a delay of 0.2 s; The operating voltage of the wheel involved in the accident is 0.8 pu, and the delay is 0.1 s. Utilizing the pre-configured emergency reserve capacity for various power generation types, voltage regulation is performed based on a multi-round voltage adjustment scheme, including: If the voltage of the power system drops to 0.84 pu, initiate voltage adjustment corresponding to the first basic cycle of photovoltaic power and continue for 15-20 seconds; During the first round of voltage regulation, a portion of the photovoltaic emergency reserve capacity is injected into the grid; If the voltage adjustment effect of the first basic cycle does not achieve the expected effect, when the voltage of the power system drops to 0.83 pu, the voltage adjustment of the second basic cycle corresponding to hydropower is initiated and lasts for 0.2 seconds; during the voltage adjustment of the second basic cycle, a portion of the emergency reserve capacity of hydropower is injected into the power grid. If the voltage adjustment effect of the second basic round does not achieve the expected effect, when the voltage of the power system drops to 0.82 pu, the voltage adjustment of the third basic round corresponding to nuclear power is initiated and lasts for 0.2 s; during the voltage adjustment of the third basic round, a portion of the emergency reserve capacity of nuclear power is injected into the power grid. If the voltage adjustment effect of the third basic wheel does not achieve the expected effect, when the voltage of the power system drops to 0.81pu, the voltage adjustment of the fourth basic wheel corresponding to thermal power is initiated and lasts for 0.2s; during the voltage adjustment of the fourth basic wheel, a portion of the emergency reserve capacity of thermal power is injected into the power grid. If the voltage adjustment effect of the fourth basic round does not achieve the expected effect, and the voltage of the power system is monitored to be as low as 0.8 pu, the voltage adjustment of the emergency round is initiated, the load is disconnected or the energy storage system is activated; during the voltage adjustment of the emergency round, all or part of the remaining emergency reserve capacity of each generation type is injected into the power grid. The method according to claim 33 is characterized in that, The frequency multi-round adjustment range includes: thermal power, hydropower, nuclear power, and energy storage; the frequency multi-round adjustment priority information includes: thermal power > hydropower > nuclear power > energy storage; The multiple frequency adjustments include: frequency adjustments on three basic cycles corresponding to thermal power, hydropower, and nuclear power, and frequency adjustments on one accident cycle corresponding to fault conditions; The frequency multi-round adjustment method includes: the operating frequency and delay of the first basic wheel, the operating frequency level and delay of the remaining basic wheels, and the operating frequency and delay of the accident wheel. The method according to claim 36, characterized in that, The operating frequency of the first basic wheel is 49.8Hz, with a delay of 0.2s; The frequency difference between the remaining basic wheels is 0.2Hz, with a delay of 0.2s; The operating frequency of the wheel involved in the accident was 49.2 Hz with a delay of 0.1 s. Utilizing the pre-configured emergency reserve capacity for various power generation types, frequency adjustment is performed based on a multi-round frequency adjustment scheme, including: If the frequency of the power system drops to 49.8Hz, the frequency adjustment corresponding to the first basic wheel of the thermal power plant is initiated and lasts for 0.2s; During the first round of frequency adjustments, a portion of the emergency reserve capacity of thermal power plants will be injected into the power grid. If the frequency adjustment of the first basic cycle does not achieve the expected results, when the frequency of the power system drops to 49.6Hz, the frequency adjustment of the second basic cycle corresponding to hydropower will be initiated and last for 0.2s; during the frequency adjustment of the second basic cycle, a portion of the emergency reserve capacity of hydropower will be injected into the power grid; If the frequency adjustment of the second basic round does not achieve the expected results, once the frequency of the power system drops to 49.4Hz, the frequency adjustment of the third basic round corresponding to nuclear power will be initiated and last for 0.2s; during the frequency adjustment of the third basic round, a portion of the emergency reserve capacity of nuclear power will be injected into the power grid; If the frequency adjustment of the third basic round does not achieve the expected results, the frequency adjustment of the emergency round will be initiated when the frequency of the power system drops to 49.2Hz, and the load will be cut off or the energy storage system will be activated. During the frequency adjustment of the emergency round, all or part of the remaining emergency reserve capacity of each generation type will be injected into the power grid. The method according to claim 33 is characterized in that, Based on the stability control requirements of the frequency dynamic characteristics of the entire power system, the projected total network load, and the characteristics of generating units at each power source, the power grid generates output curves for generating units at each power source in future time periods, including: Determine the target values ​​for frequency dynamic characteristic indicators; Based on the target value, the monitored frequency disturbance of the power system, the expected total network load, and the characteristics of the generating units at each power source, the output curves of the generating units at each power source in the future are generated. The power output curves are used to control the power of each generating unit so that the deviation between the actual value of the frequency dynamic characteristic index and the target value is within a preset index fluctuation range. The generating unit characteristics include the amount of power change that the generating unit at the power source needs to provide for a unit change in frequency. The method according to claim 38, characterized in that, Based on the target value, the monitored frequency disturbances of the power system, the projected total network load, and the unit characteristics at each power source are used to generate the power output curves for each power source in the future time period, including: Based on the target value, the monitored frequency disturbance of the power system, and the unit characteristics of each power source, the power change that each power source of each generation type should provide in order to keep the deviation between the actual value of the frequency dynamic characteristic index and the target value within the preset index fluctuation range is determined. Based on the projected total grid load for the future period and the determined power variation that each power source should provide, the output curves of the units at each power source are generated for the future period. The method according to any one of claims 33-39 is characterized in that, The determination of whether the current power system conforms to the characteristics of a large-scale power generation system and a small-scale grid based on the output of the power sources of the various power generation types includes: If the combined output of any two generators at any power source of any given power generation type exceeds 15% of the total capacity of the current power system, then the power system is deemed to conform to the characteristics of a large generator and a small grid. The method according to claim 5, characterized in that, The power function requirement for power grid self-regulation specifically refers to adjusting the power grid structure. The functional operator is a power grid structure adjustment operator, which is used to adjust the grid structure of the power grid. The process by which the digital power system utilizes an intelligent agent constructed based on the power grid structure regulation operator includes: Get the current system inertia corresponding to the current mesh structure; The frequency change rate is calculated based on the current system inertia and the difference between the power demand and the actual power supply. If the frequency change rate is greater than a preset first threshold, then the target mesh structure and the target setpoint corresponding to the target mesh structure are determined based on a pre-established deep learning model; the frequency change rate of the target mesh structure is less than or equal to the first threshold. Based on the target setpoint, modify the current setpoint applied in the current grid structure, and adjust the current grid structure of the power grid to the target grid structure. The method according to claim 41, characterized in that, The target mesh structure is determined based on a pre-established deep learning model, including: Based on a pre-established deep learning model, the inertia of the simulated system with all available mesh structures is calculated in parallel, and the available mesh structure with the largest simulated system inertia is determined as the target mesh structure. The method according to claim 41, characterized in that, The target mesh structure is determined based on a pre-established deep learning model, including: Based on the operating parameters of all selectable grid structures, a pre-established deep learning model is used to calculate the frequency change rate and the change in the difference between power generation and power demand for each of the multiple adjustment schemes that switch from the current grid structure to each selectable grid structure, thereby obtaining the scheme type to which each of the multiple adjustment schemes belongs; the scheme type includes stable type and aggressive type. If a stable adjustment scheme exists, the target mesh structure is determined from the available mesh structures corresponding to the stable adjustment scheme. The method according to claim 41, characterized in that, Before determining the target mesh structure and the target setpoint corresponding to the target mesh structure based on the pre-established deep learning model, the method further includes: If a faulty node exists, the faulty node is removed from the current grid structure; the node includes at least one of power plants, substations, and dispatch control centers. The method according to claim 41, characterized in that, The target mesh structure is determined based on a pre-established deep learning model, including: Based on a pre-established deep learning model, a target grid structure that satisfies the balance between power generation and power demand is determined according to the matching situation of power generation and power demand in the current grid structure. The method according to claim 45, characterized in that, The method, based on a pre-established deep learning model, determines a target grid structure that satisfies the balance between power generation and demand according to the matching of power generation and demand in the current grid structure, including: Obtain the rate of change of power generation of each power generation type grid node in the current grid structure; If there are grid nodes where power generation decreases and the rate of change in power generation exceeds a preset rate of change threshold, then based on the pre-established deep learning model and according to the matching situation between power generation and electricity demand in the current grid structure, a target grid structure for adding a power generation grid is determined on the basis of the current grid structure. If there are grid nodes where power generation increases and the rate of change in power generation exceeds a preset rate of change threshold, then based on the pre-established deep learning model and according to the matching situation between power generation and electricity demand in the current grid structure, a target grid structure for adding a load-type power grid is determined on the basis of the current grid structure. The method according to claim 41, characterized in that, The target mesh structure is determined based on a pre-established deep learning model, including: Based on a pre-established deep learning model, a target network structure with stability higher than a preset limit is determined according to the security priority of the current mesh structure. The method according to claim 41, characterized in that, The target mesh structure is determined based on a pre-established deep learning model, including: Based on a pre-established deep learning model and grid transfer function, the performance impact of multiple adjustment schemes that switch from the current grid structure to various optional grid structures on other grids in the power system is obtained. The target mesh structure is determined based on the pre-established deep learning model and the performance impact. The method according to claim 41, characterized in that, The target setpoints corresponding to the target mesh structure are determined based on a pre-established deep learning model, including: Using a pre-established deep learning model, based on a typical value library and the target grid structure, the target set value corresponding to the target grid structure is determined; the typical value library includes at least the correspondence between historical set value data and grid structures. The method according to claim 41, characterized in that, Adjusting the current grid structure of the power grid to the target grid structure includes: Based on the current grid structure and the target grid structure, multiple target actions that need to be performed by relays are determined; the target actions are opening or closing. Based on the action time information of each relay, multiple relays are controlled to perform target actions, thereby realizing the adjustment from the current grid structure to the target grid structure; the action time information includes the transmission time of the target setpoint, the time required to write the target setpoint, the duration of the protection action, the time required for the relay to close, the time required for the relay to open, and the time required to return status information. The method according to claim 41, characterized in that, Adjusting the current grid structure of the power grid to the target grid structure includes: Parallel computation is performed based on the target setpoints stored in a binary tree structure to obtain the action information of multiple relays that need to perform target actions during the process of adjusting the current mesh structure to the target mesh structure; the target action is opening or closing; the action information includes the time node when the relay is allowed to act and the order in which the multiple relays perform their actions; Based on the action information of each of the multiple relays, the multiple relays are controlled to perform the target action in order to adjust the current mesh structure to the target mesh structure. A device for constructing a digital power system based on a multifunctional intelligent agent, characterized in that, include: The data extraction module is used to extract characteristic production factor data corresponding to the characteristic production factor group from the power data on the power plant side; In addition, extract control production factor data corresponding to the control production factor group from the power grid side's power data; The characteristic production factor group includes characteristic production factors at multiple levels; the control production factor group includes control production factors from multiple aspects. The vector construction module is used to construct a target data vector based on the power function requirements of the digital power system using the extracted characteristic production factor data. The data determination module is used to determine the key control production element data associated with the power function requirements from the extracted control production element data; The operator construction module is used to learn the mapping function between the target data vector and the key control production factor data using artificial intelligence algorithms based on the target data vector and the key control production factor data, construct a functional operator with the mapping function as the kernel, and construct an intelligent agent based on the functional operator. The intelligent agent adding module is used to add intelligent agents corresponding to various different power function requirements to the digital power system, and to use each intelligent agent as a power function implementation unit in the digital power system. The apparatus according to claim 52 is characterized in that, The power function requirements are as follows: The power function requirements for power transactions between the grid and power plants, the power function requirements for grid-side dispatch and control of power plants, or the power function requirements for grid self-regulation. The apparatus according to claim 53 is characterized in that, The specific power function requirement for power trading between the grid side and the power plant side is to calculate electricity prices. The functional operator is an electricity price operator, used to calculate electricity prices. The process by which the digital power system utilizes an intelligent agent built based on the electricity price operator includes: The intelligent agent is invoked to obtain the electricity price calculation result on the power plant side through the electricity price operator; The electricity price calculation results are verified using a financial verification model on the grid side. The financial verification model includes: a first verification condition and a second verification condition. The first verification condition is: the actual output of various types of power plants is greater than or equal to the grid power volume. The second verification condition is: the sum of the products of the actual output of various types of power plants, the grid electricity price at the corresponding time, and the power generation duration is less than or equal to the product of the grid average price and the grid power volume. If the electricity price calculation result meets both the first verification condition and the second verification condition, then the electricity price calculation result is determined to have passed verification; if the electricity price calculation result does not meet either the first verification condition or the second verification condition, then the electricity price calculation result is determined to have failed verification. If the electricity price calculation result is verified, then a suitable and feasible electricity price calculation function for the grid side is selected from the scheduling strategy library, and the electricity price calculation function is loaded into the electricity price operator to update the electricity price operator. The updated electricity price operator is used to obtain the electricity price calculation results on the grid side. The apparatus according to claim 53 is characterized in that, The power function requirement for power trading between the grid side and the power plant side is specifically power clearing, and the functional operator is a clearing operator, which is used for power clearing; the process by which the digital power system utilizes an intelligent agent built based on the clearing operator includes: Based on the power generation element characteristics of each associated power plant in the power grid, low-level data abstraction processing is performed to obtain the first data vector of each power generation element characteristic. Long-term transaction electricity prices are predicted based on a preset long-term electricity price function library and each of the first data vectors, so as to establish a long-term transaction stack. A spot trading stack is established based on the spot trading electricity prices of each of the associated power plants, and a first clearing function is determined through the long-term trading stack and the spot trading stack. Based on the first clearing function, determine the power supply and demand balance state of the power grid when it is cleared under the first clearing function; The first clearing function is adjusted according to the power supply and demand balance state to obtain the second clearing function, and power clearing is performed based on the second clearing function. The apparatus according to claim 53 is characterized in that, The power function requirement for realizing grid-side dispatch control over power plants specifically refers to peak-shaving dispatch in high-frequency load change scenarios for thermal power plants. The functional operator is a thermal power peak-shaving dispatch operator, which is used to realize peak-shaving dispatch in high-frequency load change scenarios for thermal power plants. The process by which the digital power system utilizes an intelligent agent constructed based on the thermal power peak-shaving dispatch operator includes: Calculate the heat storage margin of energy storage devices in a thermal power plant; the heat storage margin includes a heat storage margin characterization value and a heat release capacity characterization value; The system sends the correlation data between the load, performance, and cost of each thermal power unit in the thermal power plant, as well as the heat storage margin, to the grid side so that the grid side can determine the target units for expected auxiliary peak shaving and issue peak shaving dispatch instructions based on the changes in thermal power load demand, the correlation data provided by each thermal power plant, and the heat storage margin. The system receives a peak-shaving dispatch instruction from the power grid side; the peak-shaving dispatch instruction carries the unit identifier of the target unit and peak-shaving requirement information for the target unit; the peak-shaving requirement information includes the peak-shaving load curve. Assisted peak shaving services are performed based on the peak shaving requirement information. The apparatus according to claim 53 is characterized in that, The power function requirement for realizing grid-side dispatch control over power plant side specifically refers to water resource dispatching of cascade hydropower stations. The functional operator is a water resource dispatching operator, which is used to realize water resource dispatching of cascade hydropower stations. The process by which the digital power system utilizes an intelligent agent constructed based on the water resource dispatching operator includes: Obtain water resource data for each hydropower station in the cascade hydropower station system at the current moment; For each of the aforementioned hydropower stations, based on the water resource data of that hydropower station, it is determined whether the hydropower station is in a safe operating state; When it is determined that all the hydropower stations are in the safe operating state, it is determined whether to carry out water resource scheduling based on the water resource data of each hydropower station, the preset electricity price scheme and the target historical data; the target historical data includes the average power generation water consumption and the average power plant revenue corresponding to the month in which the current moment is located. If water resource scheduling is determined, a water resource scheduling scheme is determined based on the water resource data and water resource scheduling model of each hydropower station; the water resource scheduling model is a pre-trained model used to output the water resource scheduling scheme; the water resource scheduling scheme includes day-ahead scheduling and seasonal scheduling. The apparatus according to claim 53 is characterized in that, The power function requirement for realizing grid-side dispatch control over power plants specifically refers to grid-side dispatch of power plants in a large-scale power system with small-scale grids. The functional operator is a large-scale grid dispatch operator, which is used to realize grid-side dispatch of power plants in the large-scale power system. The process of the digital power system utilizing an intelligent agent built based on the large-scale grid dispatch operator includes: Based on the digital feature extraction of the power grid and power sources of various power generation types in the power system, the emergency reserve capacity of various power generation types that should be reserved in the power system is determined and the emergency reserve capacity is configured. Based on the output of the power sources of the various power generation types mentioned above, determine whether the current power system conforms to the characteristics of large generators and small grids; If it is determined that the current power system conforms to the characteristics of a large generator and a small grid, then the power grid utilizes the configured emergency reserve capacity for various power generation types to schedule power sources within the power system for voltage adjustment based on a multi-round voltage adjustment scheme, and / or schedules power sources within the power system for frequency adjustment based on a multi-round frequency adjustment scheme. The multi-round voltage adjustment scheme includes: voltage adjustment methods, voltage adjustment ranges, and voltage adjustment priority information for the various power generation types; the multi-round frequency adjustment scheme includes: frequency adjustment methods, frequency adjustment ranges, and frequency adjustment priority information for the various power generation types. Based on the stability control requirements of the frequency dynamic characteristic index of the entire power system, the expected load of the entire network and the characteristics of the generating units at each power source, the power grid generates the output curves of the generating units at each power source in the future period; the frequency dynamic characteristic index is used to numerically characterize the amount of power change of the entire network required to cause a unit frequency change in the power system. The power grid sends generation dispatch instructions to each power source to adjust power output; the generation dispatch instructions include the output curves of the corresponding power source units. The apparatus according to claim 53 is characterized in that, The power function requirement for power grid self-regulation specifically refers to adjusting the power grid structure. The functional operator is a power grid structure adjustment operator, which is used to adjust the grid structure of the power grid. The process by which the digital power system utilizes an intelligent agent constructed based on the power grid structure regulation operator includes: Get the current system inertia corresponding to the current mesh structure; The frequency change rate is calculated based on the current system inertia and the difference between the power demand and the actual power supply. If the frequency change rate is greater than a preset first threshold, then the target mesh structure and the target setpoint corresponding to the target mesh structure are determined based on a pre-established deep learning model; the frequency change rate of the target mesh structure is less than or equal to the first threshold. Based on the target setpoint, modify the current setpoint applied in the current grid structure, and adjust the current grid structure of the power grid to the target grid structure.

Images