Virtual power plant energy supply and demand accurate matching method and system based on AI

Abstract

The invention relates to the technical field of power planning, provides an AI-based virtual power plant energy supply and demand accurate matching method and system, and solves the problems of low accuracy of virtual power plant energy supply and demand matching and low overall operation efficiency. The method comprises the following steps: acquiring an output data sequence of power supply equipment in a virtual power plant and an electricity price data sequence of an electricity market; dynamically adjusting the initial charging and discharging power of the energy storage equipment in the virtual power plant to obtain target charging and discharging power; carrying out fusion optimization on the load prediction model of each energy participation node in the virtual power plant by using a federated learning mode, and generating a global demand adjustment plan; generating a power dispatching plan based on the output data sequence and the electricity price data sequence; and the power dispatching plan and the global demand adjustment plan cooperatively act on energy supply and demand matching of the virtual power plant, and AI-based accurate energy supply and demand matching of the virtual power plant is realized. According to the invention, the accuracy of energy supply and demand matching of the virtual power plant and the overall operation efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of power planning technology, and in particular to a method and system for accurately matching energy supply and demand in a virtual power plant based on artificial intelligence (AI). Background Art

[0002] In AI-driven virtual power plant scenarios, accurately matching energy supply and demand faces complex technical requirements. Virtual power plants integrate multiple energy nodes, including distributed power sources, energy storage equipment, and adjustable loads. The output and demand of these nodes are affected by multiple factors, such as weather, time of day, and user behavior, resulting in high volatility and uncertainty. To achieve efficient energy utilization and stable supply, AI technology is needed to perceive the status of each node in real time and dynamically balance energy production and consumption, ensuring that user needs are met while reducing energy losses and improving overall economic efficiency. This is especially true in scenarios with a high proportion of renewable energy access and a dynamically changing electricity market, where the requirements for real-time and accurate supply and demand matching are even more stringent.

[0003] Currently, a typical solution to meet these technical requirements is a virtual power plant energy supply and demand matching solution based on a distributed AI model. In this solution, each energy node deploys a local AI model to autonomously predict local supply and demand and generate preliminary scheduling recommendations. Key parameters are then uploaded to a central node, which aggregates and analyzes them to form a global coordination strategy, which is then fed back to each node for execution, achieving a combination of distributed decision-making and centralized coordination.

[0004] However, this existing solution has significant flaws. First, the interaction between each node and the central node lacks a targeted data privacy protection mechanism, which can easily lead to the leakage of sensitive node information. Second, the central node's integrated optimization logic for each node's parameters is relatively simple and does not fully consider the characteristic differences and dynamic changes of different nodes. When some nodes experience abnormal fluctuations, the global coordination strategy lacks adaptability, making it difficult to ensure the accuracy and stability of supply and demand matching. Summary of the Invention

[0005] This application provides an AI-based method and system for precise matching of energy supply and demand in virtual power plants, which is used to solve the problems of low accuracy and overall operating efficiency of energy supply and demand matching in virtual power plants caused by lack of privacy protection, simple integration logic, and insufficient adaptability in existing technologies.

[0006] In the first aspect, this application provides an AI-based method for accurately matching energy supply and demand in a virtual power plant, including: Collect output data series of power equipment in virtual power plants and electricity price data series in the power market; Dynamically adjusting the initial charge and discharge power of the energy storage device in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain a target charge and discharge power; Based on the target charge and discharge power, a federated learning approach is used to integrate and optimize the load forecasting models of the energy participating nodes in the virtual power plant to generate a global demand regulation plan; Based on the output data sequence and the electricity price data sequence, generating a power dispatch plan using a pre-trained dispatch model; The power dispatching plan and the global demand regulation plan are coordinated in the energy supply and demand matching process of the virtual power plant to achieve accurate energy supply and demand matching of the virtual power plant based on AI.

[0007] Optionally, generating a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model includes: Inputting the output data sequence and the electricity price data sequence into a pre-trained scheduling model, and dividing the output data sequence into a plurality of data segments according to a preset time length by a division unit in the scheduling model; Determining the output value fluctuation range of each data segment by a determination unit in the scheduling model, and obtaining the output fluctuation characteristics corresponding to each data segment based on the output value fluctuation range; The calculation unit in the scheduling model calculates the electricity price data sequence to obtain a price change amplitude value, and generates a price fluctuation trend based on the price change amplitude value; The processing unit in the dispatching model associates the output fluctuation characteristics with the price fluctuation trend to obtain a correlation between the output fluctuation characteristics and the price fluctuation trend, and generates a power dispatching plan based on the correlation.

[0008] Optionally, calculating the electricity price data sequence to obtain a price change amplitude value, and generating a price fluctuation trend based on the price change amplitude value includes: Dividing the electricity price data sequence into a plurality of electricity price data intervals according to the preset time length; Comparing the electricity price data corresponding to adjacent time points within each of the electricity price data intervals to obtain a price change amplitude value within each of the electricity price data intervals; Determining the price fluctuation direction of each electricity price data interval based on the price change amplitude value of each electricity price data interval; The price change magnitude values ​​and the price fluctuation directions of all the electricity price data intervals are integrated in chronological order to generate a price fluctuation trend.

[0009] Optionally, based on the target charge and discharge power, the load forecasting models of the energy participating nodes in the virtual power plant are integrated and optimized by using a federated learning method to generate a global demand regulation plan, including: Based on the target charge and discharge power, adjusting the optimization parameters of the load forecasting model; By means of each energy participating node in the virtual power plant, according to the adjusted optimization parameters, the calculation results of the corresponding load forecasting model are updated, and based on the updated calculation results, the key parameters of the load forecasting model are extracted; The key parameters are sent to the central coordination node of the virtual power plant through each energy participating node. Based on the federated learning method, the key parameters of all load forecasting models are comprehensively adjusted by the central coordination node to generate final unified parameters applicable to the whole world. Sending the unified parameter to each energy participating node, so that each energy participating node optimizes the corresponding load forecasting model according to the unified parameter; Based on the central coordination node, the output results of all optimized load forecasting models are integrated to generate a global demand regulation plan.

[0010] Optionally, the key parameters are sent to the central coordination node of the virtual power plant through each energy participating node. Based on a federated learning approach, the key parameters of all load forecasting models are comprehensively adjusted by the central coordination node to generate final unified parameters applicable to the global load forecasting model, including: Based on the federated learning method, the central coordination node is used to determine whether the difference value of each key parameter exceeds a preset range, and the key parameter whose difference value exceeds the preset range is determined as the difference parameter; Based on the central coordination node, the difference parameters are gradient-adjusted, and the difference parameters after gradient adjustment are combined with the indifference parameters to form preliminary unified parameters, where the indifference parameters are key parameters whose difference values ​​do not exceed a preset range; Distributing the preliminary unified parameters to each energy participating node, each node using a local validation data set to calculate the model performance index corresponding to the preliminary unified parameters, and feeding back the performance change between the model performance index and the original performance index to the central coordination node; The central coordination node amends the preliminary unified parameters based on the performance variation to generate final unified parameters applicable to the global situation.

[0011] Optionally, the coordinating the power dispatch plan and the global demand regulation plan in the energy supply and demand matching process of the virtual power plant includes: According to the preset time length, the power dispatch plan and the global demand adjustment plan are divided respectively to obtain a plurality of time units corresponding to each; Matching the time units of the power dispatch plan with the time units of the global demand adjustment plan in chronological order, so that the plan contents of both parties in each time unit group form a time association, and obtaining a time association plan corresponding to each time unit group; Based on the time association plan corresponding to the same time unit group, identifying the inconsistent parts between the power dispatch plan and the global demand adjustment plan in the same time unit as the plan conflict points; Determining adjustment rules for the plan conflict points based on real-time energy supply and demand information of the virtual power plant; Based on the adjustment rules, the plan conflict points in each of the time-related plans are corrected to form a collaborative plan, and the collaborative plan is applied to the energy supply and demand matching process of the virtual power plant.

[0012] Optionally, determining the adjustment rules for the plan conflict points based on the real-time energy supply and demand information of the virtual power plant includes: Acquire real-time energy supply and demand information of the virtual power plant, wherein the real-time energy supply and demand information includes: the real-time total energy supply and real-time total energy demand of the virtual power plant and real-time status data of energy storage equipment; Determining a real-time supply and demand state of the virtual power plant based on a comparison result of the real-time total energy supply and the real-time total energy demand, wherein the real-time supply and demand state includes: sufficient supply, balanced supply, and insufficient supply; If the real-time supply and demand status is in sufficient supply, adjusting the non-essential load demand in the global demand adjustment plan to obtain a first adjustment rule; Alternatively, if the real-time supply and demand status is in short supply, the non-critical power generation output in the power dispatch plan is adjusted to obtain a second adjustment rule; Alternatively, if the real-time supply and demand status is in supply balance, a third adjustment rule is generated based on the real-time status data of the energy storage device.

[0013] Secondly, this application provides an AI-based virtual power plant energy supply and demand precise matching system, including: The acquisition module is used to collect the output data series of the power equipment in the virtual power plant and the electricity price data series of the power market; an adjustment module, configured to dynamically adjust the initial charge and discharge power of the energy storage device in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain a target charge and discharge power; an optimization module for fusing and optimizing the load forecasting models of the energy participating nodes in the virtual power plant based on the target charge and discharge power using a federated learning approach to generate a global demand regulation plan; A generation module, configured to generate a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model; A matching module is used to coordinate the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant to achieve accurate matching of energy supply and demand of the virtual power plant based on AI.

[0014] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an AI-based virtual power plant energy supply and demand precise matching method as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements an AI-based method for accurately matching energy supply and demand of a virtual power plant as described in any one of the first aspects.

[0016] In the present application, a method for precise matching of energy supply and demand of a virtual power plant based on AI is provided, which includes: collecting the output data sequence of the power equipment in the virtual power plant and the electricity price data sequence of the power market; dynamically adjusting the initial charge and discharge power of the energy storage equipment in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain the target charge and discharge power; based on the target charge and discharge power, using a federated learning method to integrate and optimize the load forecasting model of each energy participating node in the virtual power plant to generate a global demand adjustment plan; based on the output data sequence and the electricity price data sequence, using a pre-trained scheduling model to generate a power dispatching plan; and synergizing the power dispatching plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant to achieve precise matching of energy supply and demand of the virtual power plant based on AI.

[0017] This application has the following advantages: By collecting the output data sequence of the power supply equipment in the virtual power plant and the electricity price data sequence of the power market, basic data support can be provided for subsequent charge and discharge power adjustment, scheduling plan generation, etc.; by dynamically adjusting the initial charge and discharge power of the energy storage equipment in the virtual power plant according to the output data sequence and the electricity price data sequence, the target charge and discharge power is obtained, which can make the charge and discharge power of the energy storage equipment adapt to the actual energy output and market price changes, and improve the rationality of energy storage utilization; by using the federated learning method based on the target charge and discharge power, the load forecasting model of each energy participating node in the virtual power plant is integrated and optimized. By generating a global demand adjustment plan, it is possible to improve the global adaptability of the load forecasting model while protecting the data privacy of each node, and generate an adjustment plan that is more in line with the overall demand; by generating a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model, it is possible to combine energy output fluctuations with market price changes to generate a scientific power generation and distribution strategy; by coordinating the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant, it is possible to achieve accurate matching of energy supply and demand of the virtual power plant based on AI, thereby improving energy utilization efficiency and supply stability.

[0018] Furthermore, the output data series and electricity price data series are input into the dispatch model. The output data is segmented and the output fluctuation characteristics are determined. The electricity price data is then divided into intervals, and the price change amplitude and fluctuation direction are calculated to generate the price fluctuation trend. The two are then correlated to generate a power dispatch plan. The technical effect is that by refining the time dimension of the output data and electricity price data, the output fluctuation characteristics and price fluctuation trends are accurately captured. The power dispatch plan generated by combining the correlation between the two can better align with the actual energy supply capacity of the virtual power plant and market dynamics, improving the scientific nature and adaptability of the dispatch plan.

[0019] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 A flowchart of an AI-based method for accurately matching energy supply and demand in a virtual power plant provided in an embodiment of the present application; Figure 2A schematic diagram of the structure of an AI-based virtual power plant energy supply and demand precise matching system provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0023] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0024] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0025] To address the issues of lack of privacy protection, simple integration logic, and insufficient adaptability in existing technologies, an embodiment of the present application provides an AI-based method for accurately matching energy supply and demand in virtual power plants. The method adopts the following concepts: by collecting output data of power devices and electricity price data from the power market, a basis for subsequent adjustments and plan generation is provided; by adjusting the charge and discharge power of energy storage devices based on this data, the operation of energy storage devices is more closely aligned with actual needs; by optimizing the load forecasting model of each node and generating a global demand adjustment plan through federated learning, model optimization can be achieved without directly sharing raw data between nodes, protecting information security; by combining the output and price fluctuations of the scheduling model to generate a power scheduling plan, making the plan more consistent with actual changes; and by coordinating the two plans to match energy supply and demand, overall adaptability is improved. Furthermore, the core sub-step divides the data into segments and intervals, analyzes the output fluctuation characteristics and price fluctuation trends, and generates a scheduling plan based on these, making the plan generation process more detailed and better adaptable to different situations. Through these steps, the privacy protection issue in information exchange is resolved, and the plan generation and integration are more realistic, improving the adaptability to dynamic changes, thereby achieving accurate matching of energy supply and demand.

[0026] Figure 1 A flowchart of an AI-based virtual power plant energy supply and demand precise matching method provided in an embodiment of the present application, such as Figure 1 As shown, the method includes: S11. Collect the output data series of the power equipment in the virtual power plant and the electricity price data series of the power market.

[0027] Among them, the virtual power plant is an energy system that integrates multiple decentralized energy devices such as solar energy and wind energy and operates them in a unified and coordinated manner. The power supply equipment is the equipment that generates electricity in the system, such as solar panels and wind turbines. The output data sequence is a string of data arranged in chronological order on the power generation of these devices at different time points, including the specific power generation values ​​at each time point. The power market is a place for power trading, and its electricity price will fluctuate with changes in supply and demand. The electricity price data sequence is a string of data on the electricity prices at different time points in the power market, arranged in chronological order, including the specific prices at each time point. The result of this step is to obtain the above two types of data sequences to provide basic information for subsequent steps.

[0028] In an embodiment of the present application, a monitoring device installed on the power supply equipment records the power generation of the equipment at fixed time intervals (such as every hour), and these data are sorted in chronological order to form an output data sequence. Then, the electricity prices of the same time intervals are obtained from the information platform of the electricity market and sorted in chronological order to form an electricity price data sequence. For example, the solar panels of the virtual power plant A generate 100 degrees of electricity at 8 am and 120 degrees of electricity at 9 am, and the wind turbines generate 50 degrees of electricity at 8 am and 40 degrees of electricity at 9 am. The output data sequences are sorted in chronological order as follows: 100 degrees of solar energy and 50 degrees of wind energy at 8 am, and 120 degrees of solar energy and 40 degrees of wind energy at 9 am. At the same time, the electricity prices of 0.5 yuan / degree at 8 am and 0.55 yuan / degree at 9 am are obtained from the electricity market and sorted as the electricity price data sequences of 0.5 yuan / degree at 8 am and 0.55 yuan / degree at 9 am.

[0029] S12. Dynamically adjust the initial charge and discharge power of the energy storage equipment in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain the target charge and discharge power.

[0030] Among them, energy storage devices are devices that can store electricity (charging) and release electricity (discharging), such as battery packs. The initial charge and discharge power is the amount of electricity charged or discharged by the energy storage device per unit time before adjustment. The target charge and discharge power is the charge and discharge power that is adjusted according to the output data sequence and electricity price data sequence to suit the current actual situation. The result generated in this step is the target charge and discharge power that adapts to actual needs.

[0031] In an embodiment of the present application, the output data sequence is first analyzed to determine whether the power generation of the power supply equipment is sufficient. If the power generation is large in a certain period and there is excess power to be stored, the charging power is increased. The electricity price data sequence is then combined to determine the electricity price. If the electricity price is low in a certain period and it is suitable for more charging, the extent of increasing the charging power is further determined. If the electricity price is high and it is suitable for more discharging, the discharge power is adjusted. Through the above analysis, the initial charging and discharging power is dynamically adjusted to obtain the target charging and discharging power. For example, the virtual power plant A generates 200 kWh of solar energy per hour from 12 noon to 14 o'clock, and the demand is only 100 kWh, with 100 kWh of excess power. At this time, the electricity price is low at 0.4 yuan / kWh. It is calculated that the excess power can support the charging power from the initial 40kW to 90kW, so the charging power is adjusted to 90kW.

[0032] S13. Based on the target charging and discharging power, the load forecasting models of each energy participating node in the virtual power plant are integrated and optimized using federated learning to generate a global demand regulation plan.

[0033] Among them, the federated learning method refers to a method in which multiple participating units (such as energy nodes) jointly optimize a model without directly sharing the original data. The energy participating nodes are units that use or regulate energy in the virtual power plant, such as residential areas, factories, shopping malls, etc. The load forecasting model is a model used to predict the electricity consumption of each node in the future. The global demand adjustment plan is an overall plan to coordinate the electricity demand of each node, including the electricity consumption period and electricity distribution of each node. The result generated in this step is to generate the global plan through the optimization model.

[0034] In an embodiment of the present application, each energy participating node first uses local electricity consumption data and the target charge and discharge power obtained by S12 to perform preliminary optimization of its own load forecasting model locally. This process does not share the original data and only adjusts the parameters in the model that affect the prediction. The optimized model parameters (such as calculation coefficients) are then sent to the central coordination unit. The central coordination unit summarizes these parameters (such as calculating the average value) to obtain global model parameters and sends them to each node. Each node updates the model with the global parameters and optimizes it again. The above process is repeated until the model is accurate. Finally, based on the optimized model, the electricity consumption of each node is predicted and integrated to generate a global demand adjustment plan. For example, residential area B uses local electricity consumption records and target charge and discharge power to optimize the model and uploads parameter 0.3, factory C uploads parameter 0.4, and the center calculates the average value of 0.35 and sends it. After both parties update the model, it predicts that B needs 200 degrees / day and C needs 500 degrees / day. It is planned that B will use electricity in the morning and C will use electricity in the afternoon.

[0035] S14. Based on the output data sequence and the electricity price data sequence, a power dispatch plan is generated using a pre-trained dispatch model.

[0036] Among them, the scheduling model is a model that has been pre-trained with historical data and can generate a power distribution plan based on the input data. The power scheduling plan is a plan to guide the generation and distribution of electricity in the virtual power plant, including the power generation, storage capacity and supply objects in each time period. The result of this step is to generate the scheduling plan through the model by combining the output and electricity price data.

[0037] In an embodiment of the present application, the output data sequence and electricity price data sequence obtained by S11 are input into a pre-trained scheduling model. The model first analyzes the fluctuation characteristics of the output data to determine which periods generate more electricity and which periods generate less electricity, and then analyzes the fluctuation trend of the electricity price data to determine which periods have high electricity prices and which periods have low electricity prices. Then, the output fluctuation and the electricity price fluctuation are correlated (such as storing more electricity when more electricity is generated and the electricity price is low, and giving priority to supplying high-demand nodes when more electricity is generated and the electricity price is high). Based on the above analysis, a power scheduling plan is generated. For example, the model analysis finds that the wind power generation of virtual power plant A is 150 degrees per hour (more) from 2 am to 6 am and the electricity price is 0.3 yuan / degree (lower), and it is planned to store more electricity; the solar energy is 200 degrees per hour (more) from 12 noon to 14 o'clock and the electricity price is 0.6 yuan / degree (higher), and it is planned to give priority to supplying shopping malls.

[0038] S15. Coordinate the power dispatch plan and the global demand regulation plan in the energy supply and demand matching process of the virtual power plant to achieve accurate energy supply and demand matching of the virtual power plant based on AI.

[0039] Among them, synergy refers to combining the power dispatch plan and the global demand adjustment plan to jointly guide the energy supply and demand matching. The energy supply and demand matching process is the process of making the power generation of the virtual power plant and the power consumption of each node adapt to each other. Precise matching means that the power generation just meets the power consumption without waste or shortage. The result of this step is to achieve precise matching of energy supply and demand of the virtual power plant based on AI.

[0040] In an embodiment of the present application, the power dispatch plan generated by S14 is compared with the global demand adjustment plan generated by S13 to check whether the power generation recommendation and power consumption demand in the same period are consistent. If there is a difference, the difference is adjusted according to the real-time situation of the virtual power plant (such as the status of the energy storage device). Finally, the adjusted collaborative plan is applied to the energy supply and demand matching process to achieve accurate matching. For example, the dispatch plan of virtual power plant A recommends generating 700 degrees of power at 3 pm. The demand plan shows that the total demand is 650 degrees, with a difference of 50 degrees. After checking the real-time situation, it is found that the energy storage device still has 50 degrees of storage space, so it is adjusted to generate 650 degrees of power to just meet the demand.

[0041] This application provides the following specific example: Virtual Power Plant A includes solar and wind power equipment, energy storage equipment, and participating energy nodes such as Residential Area B, Factory C, and Shopping Mall D. Staff members record power generation and electricity market prices at different points in time and organize them into a chronological data sequence. Based on this data, they analyze the sufficiency of power generation and the level of electricity prices during different time periods, and adjust the charge and discharge power of the energy storage equipment to better reflect actual conditions. Each participating energy node uses local power consumption data and the adjusted charge and discharge power to locally optimize its own power consumption forecast model. The model parameters are then sent to a central coordination unit, which integrates and distributes the optimized parameters. After each node updates its model, it generates a global demand plan that coordinates its power consumption during each time period. Simultaneously, combining power generation and electricity price data, a scheduling model is used to generate a power dispatch plan that specifies power storage and supply arrangements for different time periods. Finally, the two plans are compared and adjusted based on real-time conditions such as the status of the energy storage equipment to ensure that power generation just meets the needs of each node, achieving precise matching of energy supply and demand.

[0042] By executing S11~S15, the embodiment of the present application provides accurate basic information for the entire process by collecting power generation data of power supply equipment and electricity market price data, ensuring that all adjustments and plans are based on actual conditions. Dynamically adjust the charging and discharging power of energy storage equipment so that its operation is more in line with changes in power generation and electricity prices, reduce power waste and improve economy, and lay a reasonable energy storage foundation for subsequent steps. Federated learning is used to optimize the load forecasting model to improve forecast accuracy while protecting data privacy. The generated global demand adjustment plan can effectively coordinate the power consumption of each node and avoid local power concentration. The scheduling model combines power generation capacity and price changes to generate a power scheduling plan, making the generation, storage and distribution of electricity more reasonable, taking into account both supply stability and economy. Ultimately, by coordinating the two plans, the power generation and power consumption are dynamically adapted, achieving a precise match between energy supply and demand of the virtual power plant, and improving the overall energy utilization efficiency and supply stability.

[0043] In a possible embodiment, S14, based on the output data sequence and the electricity price data sequence, using a pre-trained scheduling model to generate a power scheduling plan, includes: Step 141: Input the output data sequence and the electricity price data sequence into a pre-trained scheduling model, and divide the output data sequence into multiple data segments according to preset time lengths through a division unit in the scheduling model.

[0044] Among them, the output data sequence is a data string of the power generation of the power equipment at different time points arranged in time, the electricity price data sequence is a data string of the electricity prices at different time points in the power market arranged in time, the pre-trained scheduling model is a model trained in advance with historical data and can process data to generate power plans, the division unit is the part of the scheduling model that divides the data, the preset time length is a fixed time set in advance (such as a few hours), and the data segment is a section of power generation data divided according to the preset time. The result generated in this step is to divide the output data sequence into multiple data segments.

[0045] In an embodiment of the present application, the output data sequence and the electricity price data sequence are input into a pre-trained scheduling model. The division unit in the model divides the output data sequence into multiple continuous data segments according to a preset time length (such as 4 hours). For example, the output data of a virtual power plant for 24 hours a day is divided into 4 hours, 24÷4=6, and is divided into 6 data segments, each containing 4 hours of power generation data.

[0046] Step 142: Determine the output value fluctuation range of each data segment through the determination unit in the scheduling model, and obtain the output fluctuation characteristics corresponding to each data segment based on the output value fluctuation range.

[0047] Among them, the determination unit is the part of the scheduling model that determines the fluctuation range of power generation. The output value fluctuation range is the difference between the maximum and minimum power generation values ​​in each data segment. The output fluctuation characteristics are the characteristics that reflect the size of the power generation change within the data segment (such as large or small fluctuations). The result generated in this step is the output fluctuation characteristics corresponding to each data segment.

[0048] In an embodiment of the present application, the determination unit of the scheduling model finds the maximum and minimum values ​​of the power generation in each data segment, uses the maximum value minus the minimum value to calculate the output value fluctuation range, and then determines the output fluctuation characteristics (such as large or small fluctuations) based on the range size. For example, a data segment has a maximum of 80 degrees and a minimum of 50 degrees, 80-50=30 degrees, and the characteristic is "medium fluctuation".

[0049] Step 143: Calculate the electricity price data sequence through the calculation unit in the dispatching model to obtain the price change amplitude value, and generate the price fluctuation trend based on the price change amplitude value.

[0050] Among them, the calculation unit is the part in the scheduling model that calculates the change in electricity price. The price change amplitude value is the difference between electricity prices at adjacent time points. The price fluctuation trend reflects the overall increase, decrease or stability of electricity prices over a period of time. The result generated by this step is the price fluctuation trend.

[0051] In an embodiment of the present application, the calculation unit of the scheduling model compares the electricity prices at adjacent time points in the electricity price data sequence, subtracts the electricity price at the latter time point from the price at the previous time point to obtain the price change amplitude value, integrates these values ​​to judge the overall increase, decrease or stability, and generates a price fluctuation trend. For example, the electricity price increases from 0.5 yuan to 0.6 yuan and then to 0.7 yuan, the amplitude values ​​are all positive, and the trend is "continuous increase."

[0052] Step 144: Correlation processing is performed on the output fluctuation characteristics and the price fluctuation trend by the processing unit in the dispatching model to obtain a correlation relationship between the output fluctuation characteristics and the price fluctuation trend, and a power dispatching plan is generated based on the correlation relationship.

[0053] Among them, the processing unit is the integrated analysis part of the scheduling model, the correlation processing is to analyze the relationship between the output fluctuation characteristics and the price fluctuation trend, the correlation relationship is the correspondence between the two, and the power scheduling plan is the power storage and supply arrangement formulated based on the correlation relationship. The result generated by this step is the power scheduling plan.

[0054] In an embodiment of the present application, the processing unit of the scheduling model associates the output fluctuation characteristics of each data segment with the price fluctuation trend of the corresponding time period, analyzes the correlation between the two in the same time period, and then formulates power storage and supply arrangements based on this, and generates a power scheduling plan. For example, "electricity prices rise when power generation fluctuations are large", and the plan prioritizes supply to high-demand nodes and reduces storage.

[0055] This application provides the following specific example: Virtual power plant A inputs a 24-hour daily output data sequence and electricity price data sequence into a scheduling model. The model's partitioning unit divides the output data into eight data segments (24 ÷ 3 = 8), each with a preset 3-hour timeframe. The determination unit analyzes each data segment. For example, in one 3-hour segment, the maximum power generation is 120 kWh and the minimum is 60 kWh, where 120 - 60 = 60 kWh, indicating "large fluctuations." Another segment has a maximum of 70 kWh and a minimum of 40 kWh, where 70 - 40 = 30 kWh, indicating "small fluctuations." The calculation unit processes the electricity price data and finds that the price of electricity at adjacent time points ranges from 0.5 yuan to 0.55 yuan (a 0.05 yuan increment), and then to 0.52 yuan (a -0.03 yuan increment), generating a trend of "first increasing, then slightly decreasing." The processing unit correlates and discovers that periods of large fluctuations correspond to periods of initial price increases followed by slight decreases. Based on this, a plan is generated: when fluctuations are large and prices are rising, supply is prioritized to demand nodes such as shopping malls; when fluctuations are small, more power is stored.

[0056] By executing steps 141 through 144, this embodiment of the present application makes power generation analysis more targeted by segmenting output data by time; determines output fluctuation characteristics and price fluctuation trends, clearly presenting the changing patterns of power generation and electricity prices; and correlates these two to generate a scheduling plan, allowing power storage and supply to take into account both power generation fluctuations and electricity price changes, avoiding irrational plans caused by a single factor. This overall process improves the orderliness of data processing, makes power scheduling plans more realistic, improves energy utilization efficiency, and provides a scientific basis for accurately matching energy supply and demand in virtual power plants.

[0057] In a possible embodiment, step 143 of calculating the electricity price data sequence to obtain a price change amplitude value, and generating a price fluctuation trend based on the price change amplitude value, includes: a1. Divide the electricity price data sequence into multiple electricity price data intervals according to a preset time length.

[0058] Among them, the electricity price data sequence is a string of data in chronological order of electricity prices at different time points in the electricity market. The preset time length is a fixed period of time set in advance (such as a few hours). The electricity price data interval is an interval obtained by dividing according to the preset time length and containing the electricity price data of the time period. The result of this step is to divide the electricity price data sequence into multiple electricity price data intervals.

[0059] In an embodiment of the present application, a preset time length (such as 4 hours) is determined, and the electricity price data sequence is divided according to this time length into multiple non-overlapping electricity price data intervals, each interval containing electricity price data at all time points within the time period. For example, the electricity price data sequence of a virtual power plant records the electricity price for 24 hours a day, and is divided according to the preset time length of 4 hours. 24 divided by 4 equals 6, so it is divided into 6 electricity price data intervals, each interval containing 4 hours of electricity price data.

[0060] a2. Compare the electricity price data corresponding to adjacent time points within each electricity price data interval to obtain the price change amplitude value within each electricity price data interval.

[0061] Among them, the electricity price data interval is a section of electricity price data divided according to a preset time length, adjacent time points are two consecutive time points in the same interval (such as 8 o'clock and 9 o'clock), the electricity price data is the specific electricity price corresponding to each time point, and the price change amplitude value is the difference between the electricity prices at adjacent time points in the same interval. The result generated in this step is the price change amplitude value within each electricity price data interval.

[0062] In an embodiment of the present application, for each electricity price data interval, all adjacent time points in the interval are found, the electricity price data of these adjacent time points are compared, and the electricity price of the latter time point is subtracted from the electricity price of the previous time point to obtain the price change amplitude value of each adjacent time point, and then these amplitude values ​​in the same interval are integrated as the price change amplitude value of the interval. For example, a certain interval includes electricity prices at 8 o'clock, 9 o'clock, and 10 o'clock. The electricity price rises from 0.5 yuan to 0.55 yuan from 8 o'clock to 9 o'clock, and 0.55 minus 0.5 equals 0.05 yuan. From 9 o'clock to 10 o'clock, the electricity price drops from 0.55 yuan to 0.52 yuan, and 0.52 minus 0.55 equals -0.03 yuan. The price change amplitude values ​​of this interval are 0.05 yuan and -0.03 yuan.

[0063] a3. Based on the price change amplitude value of each electricity price data interval, determine the price fluctuation direction of each electricity price data interval.

[0064] The price change amplitude is the difference between electricity prices at adjacent time points, and the price fluctuation direction is the overall change direction of electricity prices within each electricity price data interval (such as rising, falling, or basically stable). The result generated in this step is to determine the price fluctuation direction of each electricity price data interval.

[0065] In the embodiment of the present application, the price change amplitude value within each electricity price data interval is checked. If the amplitude value is positive, the electricity price in the interval is on an overall upward trend; if it is negative, it is on a downward trend; if it is close to 0, it is basically stable. In this way, the price fluctuation direction of each interval is determined. For example, the price change amplitude value of a certain interval is 0.05 yuan and 0.03 yuan, both of which are positive values. Therefore, the price fluctuation direction of the interval is upward.

[0066] a4. Integrate the price change magnitude and price fluctuation direction of all electricity price data intervals in chronological order to generate a price fluctuation trend.

[0067] Among them, the time sequence is the order in which the electricity price data intervals are arranged in chronological order, the price change amplitude value is the difference between the electricity prices at adjacent time points, the price fluctuation direction is the change direction of the electricity prices in each interval, and the price fluctuation trend is the overall change in electricity prices over a period of time formed by integrating the amplitude values ​​and directions of all intervals. The result generated in this step is the price fluctuation trend.

[0068] In an embodiment of the present application, the electricity price data intervals are arranged in chronological order, and the price change amplitude values ​​and price fluctuation directions of all intervals are integrated to form the overall change of electricity prices over a period of time, that is, the price fluctuation trend. For example, the fluctuation directions of the six intervals are rise, rise, fall, basically stable, rise, and fall, respectively. The price fluctuation trend generated after integration is "rise first and then fall, short-term stability in the middle, rise again in the later period and then fall again."

[0069] This application provides the following specific example: The electricity price data sequence of virtual power plant A records the electricity price for 24 hours a day (one data per hour). First, it is divided according to the preset time length of 3 hours. 24 divided by 3 equals 8, and 8 electricity price data intervals are obtained, each interval contains 3 hours of electricity price data. For the first interval (8 o'clock-11 o'clock), the electricity prices at adjacent time points are compared: from 8 o'clock to 9 o'clock, it rises from 0.48 yuan to 0.52 yuan, 0.52 minus 0.48 equals 0.04 yuan; from 9 o'clock to 10 o'clock, it falls from 0.52 yuan to 0.5 yuan, 0.5 minus 0.52 equals -0.02 yuan; from 10 o'clock to 11 o'clock, it rises from 0.5 yuan to 0.55 yuan, 0.55 minus 0.5 equals 0.05 yuan, and the price change values ​​for this interval are 0.04 yuan, -0.02 yuan, and 0.05 yuan. Based on these amplitude values, the first interval shows both increases and decreases, but the increases are greater, so the direction of fluctuation is determined to be "slightly rising." The remaining seven intervals are processed in the same way to obtain their own fluctuation directions (such as "falling" and "basically stable."). Finally, the information from the eight intervals is integrated chronologically to generate the price fluctuation trend of "slightly rising in the morning, briefly falling at noon, basically stable in the afternoon, and continuing to rise in the evening."

[0070] By executing steps a1 to a4, the embodiment of the present application divides the electricity price data sequence into intervals according to fixed time periods, converting continuous electricity price data into units that are easy to analyze, allowing electricity price changes in each time period to be studied separately; calculating the price change amplitude values ​​at adjacent time points can accurately capture subtle changes in electricity prices within each interval; determining the fluctuation direction based on the amplitude value, making the electricity price change trend in each interval clearer; and finally integrating the information of all intervals to generate price fluctuation trends, fully presenting the overall change pattern of electricity prices over a period of time. The entire process makes electricity price analysis from scattered data systematic and orderly, providing a clear basis for price changes for the subsequent formulation of power dispatch plans based on power generation conditions, and helping to improve the rationality and accuracy of the plan.

[0071] In one possible embodiment, S13, based on the target charge and discharge power, a federated learning approach is used to integrate and optimize the load forecasting models of the energy participating nodes in the virtual power plant to generate a global demand adjustment plan, including: Step 131: Adjust the optimization parameters of the load forecasting model based on the target charge and discharge power.

[0072] The target charge and discharge power is the amount of power suitable for the energy storage device to charge and discharge after adjustment based on power generation and electricity prices. The load forecasting model is a model used to predict the future power consumption of each node. The optimization parameters are adjustable values ​​in the model that affect the prediction results (such as calculation coefficients). The result generated in this step is the adjusted load forecasting model optimization parameters.

[0073] In an embodiment of the present application, based on the target charge and discharge power obtained in step 12, its impact on the power consumption of each node is analyzed, and the optimization parameters related to charge and discharge in the load forecasting model are adjusted accordingly (such as the coefficient reflecting the impact of energy storage equipment in the adjustment model) to make the model more in line with the actual energy storage situation. For example, if the target charge and discharge power of a node in the virtual power plant is increased, the optimization parameters related to "energy storage replenishment power" in its load forecasting model are adjusted accordingly from 0.2 to 0.3.

[0074] Step 132: Update the calculation results of the corresponding load forecasting model according to the adjusted optimization parameters through each energy participating node in the virtual power plant, and extract the key parameters of the load forecasting model based on the updated calculation results.

[0075] Among them, the energy participating node is a unit that uses or regulates energy in the virtual power plant (such as a residential area or a factory); the adjusted optimization parameters are the model parameters obtained in step 131; the load forecasting model is a model for predicting electricity consumption; the calculation result is the predicted electricity consumption output by the model; the key parameters are the core parameters in the model that play a major role in the prediction results (such as weight values); the generated result of this step is the extracted key parameters of the load forecasting model.

[0076] In an embodiment of the present application, each energy participating node uses the optimized parameters adjusted in step 131 to update its own load forecasting model, recalculate future power consumption to obtain updated results, and then extract the key parameters that have the greatest impact on the forecast results from the model (such as the weight value that determines the peak power consumption period). For example, residential area B uses the adjusted parameters to update the model, predicts that the daily peak power consumption is in the morning, and extracts the "morning period weight" in the model as the key parameter.

[0077] Step 133: Send the key parameters to the central coordination node of the virtual power plant through each energy participating node. Based on the federated learning method, the central coordination node comprehensively adjusts the key parameters of all load forecasting models to generate final unified parameters applicable to the global situation.

[0078] Among them, key parameters are the core parameters that play a major role in the model; the central coordination node is the central unit in the virtual power plant responsible for integrating the information of each node; the federated learning method is a method in which the nodes do not share the original data, but only optimize the model through parameter collaboration; comprehensive adjustment is to summarize, average and other processing of multiple parameters; the final unified parameter is the global optimization parameter applicable to all nodes; the result generated in this step is the global final unified parameter.

[0079] In an embodiment of the present application, each energy participating node sends the extracted key parameters to the central coordination node. The central coordination node adopts a federated learning method, does not process the original data, and only makes comprehensive adjustments to all key parameters (such as calculating the average value) to generate a final unified parameter applicable to the global situation. For example, residential area B sends the key parameter 0.6, factory C sends 0.7, and shopping mall D sends 0.5. The center calculates that the sum of 0.6 plus 0.7 plus 0.5 is equal to 1.8, and 1.8 divided by 3 is equal to 0.6, resulting in a unified parameter of 0.6.

[0080] Step 134: Send the unified parameters to each energy participating node, so that each energy participating node optimizes the corresponding load forecasting model according to the unified parameters.

[0081] Among them, the unified parameters are global optimization parameters generated by the central coordination node; the energy participating nodes are units that use or adjust energy; the load forecasting model is a model for predicting electricity consumption; the result generated in this step is the load forecasting model of each node after optimization based on the unified parameters.

[0082] In an embodiment of the present application, the central coordination node sends the final unified parameters to each energy participating node, and each node replaces the corresponding parameters in its own model with the parameters to optimize the load forecasting model so that the model can not only reflect the local electricity consumption characteristics but also meet the global coordination requirements. For example, if the unified parameter is 0.6, residential area B updates the corresponding parameter in the model from 0.6 (original adjusted value) to 0.6 (unified value), and factory C updates it from 0.7 to 0.6, thereby achieving global unified optimization.

[0083] Step 135: Based on the central coordination node, the output results of all optimized load forecasting models are integrated to generate a global demand adjustment plan.

[0084] Among them, the central coordination node is the central unit responsible for integrating information; the optimized load forecasting model is the model updated with unified parameters for each node; the output result is the power consumption of each node predicted by the model; the global demand adjustment plan is the overall arrangement for coordinating the power consumption time periods and power consumption of each node; the result generated by this step is the global demand adjustment plan.

[0085] In an embodiment of the present application, the central coordination node collects the output results of the optimized load forecasting model of each energy participating node (i.e., the predicted power consumption of each node), analyzes the peak power consumption period and power amount of each node, integrates them according to the global supply and demand balance principle (such as staggering peak periods), and generates a global demand adjustment plan. For example, after integration, it is found that residential areas consume more electricity in the morning and factories consume more in the afternoon. It is planned to supply electricity to residential areas in the morning and to factories in the afternoon to avoid concentrated electricity consumption.

[0086] This application provides the following specific example: Virtual power plant A includes three energy participating nodes: residential area B, factory C, and shopping mall D. First, according to the target charging and discharging power, residential area B is charged at 90kW and factory C is discharged at 70kW. The optimization parameters of the load forecasting model of each node are adjusted: residential area B adjusts the "charging period power consumption coefficient" from 0.4 to 0.6, and factory C adjusts the "discharging period power consumption coefficient" from 0.3 to 0.5. Each node updates the model with the adjusted parameters and calculates the predicted power consumption: residential area B predicts 200 kWh of electricity consumption in the morning, factory C predicts 500 kWh of electricity consumption in the afternoon, and shopping mall D predicts 300 kWh of electricity consumption in the evening; at the same time, key parameters are extracted: residential area B is "morning weight 0.6", factory C is "afternoon weight 0.7", and shopping mall D is "evening weight 0.5". These key parameters are sent to the central coordination node, and the center uses federated learning to calculate the average value: 0.6 plus 0.7 plus 0.5 equals 1.8, 1.8 divided by 3 equals 0.6, and a unified parameter of 0.6 is generated. Each node optimized the model using a value of 0.6. Factory C adjusted its "afternoon weight" from 0.7 to 0.6, making the forecast more accurate. Finally, the central coordination node collected the output of the optimized model and integrated it to find that electricity consumption is concentrated in residential areas in the morning, in factories in the afternoon, and in shopping malls in the evening. This generated a global demand adjustment plan: Residential Area B was allocated 200 kWh of electricity in the morning, Factory C was allocated 500 kWh in the afternoon, and Shopping Mall D was allocated 300 kWh in the evening to avoid overlapping peak electricity consumption.

[0087] By executing steps 131 to 135, the embodiment of the present application adjusts the model optimization parameters according to the target charge and discharge power, so that the load forecasting model is preliminarily adapted to the actual operation of the energy storage equipment, providing a reasonable basis for subsequent forecasts. Each node updates the model and extracts key parameters, which not only retains the local power consumption characteristics, but also avoids the leakage of original data and protects information security. The central coordination node uses federated learning to generate unified parameters, achieves global optimization without obtaining original data, and balances local characteristics with overall coordination. After each node optimizes the model with unified parameters, the prediction results are more consistent, providing a reliable basis for the global plan. The global demand adjustment plan that is finally integrated and generated effectively staggers the power consumption peaks of each node, avoids imbalance between supply and demand, and improves the rationality of energy distribution and the overall operating efficiency of the virtual power plant.

[0088] In one possible embodiment, in step 133, each energy participating node sends key parameters to the central coordination node of the virtual power plant. Based on a federated learning approach, the central coordination node comprehensively adjusts the key parameters of all load forecasting models to generate final unified parameters applicable to the entire system, including: b1. Based on the federated learning method, the central coordination node is used to determine whether the difference value of each key parameter exceeds the preset range, and the key parameter whose difference value exceeds the preset range is determined as the difference parameter.

[0089] Among them, the federated learning method is a method in which each node does not share the original data but only optimizes the model through parameter collaboration. The central coordination node is the central unit in the virtual power plant that integrates the information of each node. The key parameters are the core parameters in the model that play a major role in the prediction results. The difference value is the difference between the key parameters of different nodes. The preset range is a reasonable difference range set in advance. The difference parameter is the key parameter whose difference value exceeds the preset range. The result generated in this step is to determine the difference parameter.

[0090] In an embodiment of the present application, based on the federated learning method, the central coordination node first collects the key parameters of each energy participating node, calculates the difference between the same key parameters of different nodes to obtain the difference value, and then compares these difference values ​​with the preset range set in advance. If the difference value of a key parameter exceeds the preset range, the parameter is determined as the difference parameter. For example, the same key parameters of three nodes are 0.6, 0.8, and 0.5, respectively, and the preset range is plus or minus 0.1. The calculated difference value 0.8 minus 0.6 is equal to 0.2, and 0.8 minus 0.5 is equal to 0.3. Both difference values ​​exceed the preset range, so the key parameter corresponding to 0.8 is determined as the difference parameter.

[0091] b2. Based on the central coordination node, gradient adjustment is performed on the difference parameters. The difference parameters after gradient adjustment are combined with the indifference parameters to form preliminary unified parameters. The indifference parameters are key parameters whose difference values ​​do not exceed the preset range.

[0092] Among them, the central coordination node is the central unit that integrates the information of each node, the difference parameter is the parameter determined in b1, the gradient adjustment is the process of gradually adjusting the difference parameter to make it reasonable, the indifference parameter is the key parameter whose difference value does not exceed the preset range, and the preliminary unified parameter is the parameter obtained by combining the adjusted difference parameter and the indifference parameter. The result of this step is the preliminary unified parameter.

[0093] In an embodiment of the present application, the central coordination node performs a gradient adjustment on the difference parameter determined in b1, that is, gradually reduces the gap between it and other parameters, for example, gradually adjusts the difference parameter 0.8 to 0.7, and then merges the adjusted difference parameter with the indifference parameter, that is, the key parameter whose difference value does not exceed the preset range, to form a preliminary unified parameter applicable to the global. For example, the adjusted 0.7 is merged with the indifference parameters 0.6 and 0.5, and the average value of the three is calculated as 0.6 plus 0.7 plus 0.5, which equals 1.8, and 1.8 divided by 3 equals 0.6, to obtain the preliminary unified parameter 0.6.

[0094] b3. Distribute the preliminary unified parameters to each energy participating node. Each node uses the local verification data set to calculate the model performance index corresponding to the preliminary unified parameters, and feeds back the performance change between the model performance index and the original performance index to the central coordination node.

[0095] Among them, the preliminary unified parameters are the parameters obtained in b2, the energy participating nodes are the units that use energy in the virtual power plant, such as residential areas and factories, the local verification data set is the local data of each node used to test the model, the model performance index is a standard for measuring the model prediction effect, such as accuracy, the performance change is the difference between the index corresponding to the preliminary parameters and the original index, the central coordination node receives the performance change feedback from each node, and the result generated in this step is the performance change fed back to the center.

[0096] In an embodiment of the present application, the central coordination node distributes the preliminary unified parameters obtained in b2 to each energy participating node. Each node uses its own local verification data set, that is, the local historical data used to test the model, to calculate the model performance index corresponding to the parameter, that is, the standard for measuring the model prediction effect, and then compares the index with the original performance index before adjustment to obtain the performance change, that is, the difference between the two, and feeds back the performance change to the central coordination node. For example, the original performance index of a node is 80 points, and after using the preliminary unified parameters, it is 85 points, and the performance change is 5 points. The result is fed back to the center.

[0097] b4. Based on the performance change, the central coordination node modifies the preliminary unified parameters to generate the final unified parameters applicable to the global situation.

[0098] Among them, the central coordination node is the central unit that integrates the information of each node. The performance change is the difference feedback in b3. The preliminary unified parameters are the parameters obtained in b2. The final unified parameters are the parameters that are corrected and applicable to the global environment. The result generated in this step is the final unified parameters that are applicable to the global environment.

[0099] In an embodiment of the present application, the central coordination node collects the performance changes fed back by each energy participating node and analyzes the overall trend of these changes. If the performance changes of most nodes are positive, that is, the model effect is improved, fine-tuning is performed on the basis of the preliminary unified parameters. If there is a negative change, that is, the model effect is reduced, the parameters are adjusted according to the magnitude of the change, and finally the final unified parameters applicable to the global are generated. For example, the performance of most nodes is improved by about 5 points, and only one node is reduced by 2 points. Therefore, the preliminary unified parameter of 0.6 is adjusted to 0.62 as the final unified parameter.

[0100] This application provides the following specific example: The central coordination node of the virtual power plant A collects the key parameters of residential area B, factory C, and shopping mall D, which are 0.6, 0.9, and 0.5, respectively. First, the center calculates the difference value based on the federated learning method. 0.9 minus 0.6 equals 0.3, and 0.9 minus 0.5 equals 0.3. The preset range is plus or minus 0.2. Since 0.3 is out of range, 0.9 is determined as the difference parameter. Next, the center performs a gradient adjustment on the difference parameter 0.9, first adjusting it to 0.8, then adjusting it to 0.7, and then merging it with the indifference parameters 0.6 and 0.5. The average value of the three is calculated as 0.6 plus 0.7 plus 0.5, which equals 1.8. 1.8 divided by 3 equals 0.6, and the preliminary unified parameter 0.6 is obtained. After that, the center distributes the preliminary unified parameter 0.6 to each node, and each node uses a local verification data set to test the model performance. Residential area B's original performance index was 75, but after the use, it increased to 80, a change of 5 points. Factory C's original score was 70, but after the use, it increased to 78, a change of 8 points. Shopping mall D's original score was 72, but after the use, it decreased to 70, a change of -2 points. Each node reported these performance changes to the central node. Finally, the central node analyzed the feedback and found that most nodes had improved, with only a slight decline in shopping mall D. Therefore, the initial unified parameter of 0.6 was fine-tuned to 0.62, generating the final unified parameter applicable to the entire network.

[0101] By executing b1~b4, the embodiment of the present application can correct the parameters that deviate from the reasonable range by identifying the difference parameters and performing gradient adjustment, laying the foundation for generating unified parameters. The adjusted parameters are merged with the indifference parameters to form preliminary unified parameters, realizing the transition from local parameters to global parameters. Each node tests and feeds back the performance change through local data, providing a practical basis for parameter optimization and ensuring that the parameters meet the actual conditions of each node. The final unified parameters generated based on feedback correction take into account global coordination and adapt to the local characteristics of each node, thereby improving the overall prediction effect of the model. At the same time, the original data sharing is not involved in the entire process, thus protecting information security.

[0102] In a possible embodiment, S15, coordinating the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant, includes: Step 151: Divide the power dispatching plan and the global demand adjustment plan according to preset time lengths to obtain a plurality of corresponding time units.

[0103] Among them, the preset time length is a fixed period of time set in advance (such as 3 hours), the power dispatch plan is the arrangement to guide the generation and distribution of electricity, the global demand adjustment plan is the arrangement to coordinate the electricity demand of each node, and the time unit is the unit containing the plan content of the time period obtained by dividing according to the preset time length. The result of this step is to divide the two plans into multiple time units respectively.

[0104] In an embodiment of the present application, a preset time length (such as 3 hours) is determined, and the power dispatch plan and the global demand regulation plan are divided according to this time length, and each plan is divided into multiple continuous time units, each unit containing the plan content within the time period. For example, both plans of the virtual power plant are arranged for 24 hours a day, divided by 3 hours, 24 divided by 3 equals 8, so each plan is divided into 8 time units, each unit contains 3 hours of content.

[0105] Step 152: Match the time units of the power dispatch plan with the time units of the global demand regulation plan in chronological order, so that the plan contents of both parties in each time unit group form a time association, and obtain a time association plan corresponding to each time unit group.

[0106] Among them, the time unit of the power dispatch plan is the power plan time period divided according to the preset time, the time unit of the global demand adjustment plan is the power demand time period divided according to the same time, the time sequence is the order of arrangement in chronological order, the time association is the correspondence between the two plan contents in the same time period, the time association plan is the content after the two plans in the same time unit group are associated, and the generation result of this step is the time association plan corresponding to each time unit group.

[0107] In an embodiment of the present application, each time unit of the power dispatching plan is corresponded one-to-one with the time unit of the same time in the global demand regulation plan in chronological order to form a time unit group, so that the two plan contents of the same time period are associated to form a time association plan. For example, the power dispatching unit of "8:00-11:00" is corresponded with the demand regulation unit of the same time period to form a time unit group, forming a time association plan containing the supply and demand contents of the time period.

[0108] Step 153: Based on the time association plan corresponding to the same time unit group, identify the inconsistent parts between the power dispatch plan and the global demand adjustment plan in the same time unit as the plan conflict points.

[0109] Among them, the same time unit group is the two corresponding time units in step 152, the time association plan is the plan content after association within the group, the power dispatching plan is the power supply arrangement, the global demand adjustment plan is the electricity demand arrangement, and the plan conflict point is the part where the two plan contents are inconsistent in the same time unit. The generation result of this step is the identified plan conflict point.

[0110] In an embodiment of the present application, the time-related plans corresponding to the same time unit group are checked, and the supply of the power dispatching plan within the time period is compared with the demand of the global demand adjustment plan. If the two are not equal or the allocation objects do not match, the inconsistent part is determined as a plan conflict point. For example, in a certain time unit group, the power dispatching plan arranges to supply 500 degrees, and the demand adjustment plan shows that 550 degrees are required. The difference of 50 degrees is the plan conflict point.

[0111] Step 154: Determine adjustment rules for planning conflict points based on the real-time energy supply and demand information of the virtual power plant.

[0112] Among them, the real-time energy supply and demand information of the virtual power plant is the current actual power generation, the real-time power consumption of each node, the status of the energy storage equipment and other information. The planned conflict point is the inconsistent part identified in step 153. The adjustment rule is the specific criteria used to correct the conflict point (such as giving priority to meeting demand or adjusting supply). The result generated by this step is the determined planned conflict point adjustment rule.

[0113] In an embodiment of the present application, real-time energy supply and demand information of the virtual power plant (such as the current actual power generation, the status of the energy storage equipment, etc.) is collected, and the relationship between this information and the planned conflict points is analyzed. If the real-time power generation is sufficient and there is room for energy storage, a rule of "increasing supply to meet demand" is formulated; if the power generation is insufficient but there is power in the energy storage, a rule of "releasing energy storage to supplement supply" is formulated. For example, if the conflict point is that the supply is 50 degrees less and there is a surplus in the real-time power generation, the adjustment rule is determined to increase the supply by 50 degrees.

[0114] Step 155: Based on the adjustment rules, the plan conflict points in each time-related plan are corrected to form a collaborative plan, and the collaborative plan is applied to the energy supply and demand matching process of the virtual power plant.

[0115] Among them, the adjustment rules are the criteria for correcting plan conflict points, the time-related plans are the related plan contents of the same time unit group, the plan conflict points are the inconsistent parts of the two plans, the collaborative plan is the unified plan formed after correcting the conflict points, and the energy supply and demand matching process is the process of making the power generation and power consumption adapt to each other. The result generated in this step is the collaborative plan applied to energy supply and demand matching.

[0116] In an embodiment of the present application, the plan conflict points in each time-related plan are corrected according to the adjustment rules. If the supply is insufficient, the supply is increased, and if the supply is excessive, the supply is reduced, so that the supply and demand in the same time unit are consistent. The corrected plans are integrated to form a collaborative plan, which is applied to the energy supply and demand matching process. For example, if the supply at a conflict point is 50 degrees less, the supply is increased by 50 degrees according to the rules. After correction, a collaborative plan is formed to guide the power distribution during this period.

[0117] This application provides the following specific example: Virtual Power Plant A divides the power dispatch plan and the global demand adjustment plan into 8 time units (24 ÷ 3 = 8) based on a preset time length of 3 hours. The units with the same time in the two plans are matched, such as the power dispatch unit and the demand adjustment unit from "8:00-11:00" to form a time unit group, forming a time-related plan that includes the supply and demand content for this period. A comparison shows that in the time unit group from "11:00-14:00", the power dispatch plan supplies 500 degrees, while the demand adjustment plan demands 550 degrees. The calculated difference of 50 degrees is identified as a plan conflict point. Checking real-time information shows that the current power generation is 60 degrees more than planned, and the energy storage equipment has idle space. The adjustment rule is determined to allocate 50 degrees of additional power from the excess power to supplement the supply. This rule is used to correct the conflict point, so that the supply becomes 550 degrees, which is consistent with demand. All corrected plans are integrated to form a coordinated plan, which is applied to the energy supply and demand matching process.

[0118] By executing steps 151 to 155, the embodiment of the present application divides the two plans into time units according to a preset time length, decomposes the overall plan into time periods that are easy to compare, and provides a basis for analyzing differences in the same time period. Time-related plans are formed for corresponding time units, so that the supply and demand contents of the same time period are correlated with each other, which is convenient for subsequent comparison. Identifying the conflict points of the plans clarifies the inconsistencies between the two plans and provides specific objects for adjustment. Adjustment rules are determined based on real-time information to ensure that the corrections have a practical basis and are in line with the operating status of the virtual power plant. The final collaborative plan is applied to supply and demand matching to ensure that the power supply and demand in each time period are consistent, avoid excess or shortage, achieve accurate matching of energy supply and demand of the virtual power plant, and improve overall operating efficiency.

[0119] In one possible embodiment, step 154, determining adjustment rules for plan conflict points based on real-time energy supply and demand information of the virtual power plant, includes: c1. Obtain real-time energy supply and demand information of the virtual power plant. The real-time energy supply and demand information includes: the real-time total energy supply of the virtual power plant, the real-time total energy demand and the real-time status data of the energy storage equipment.

[0120] Among them, the virtual power plant is a system that integrates distributed energy equipment, and the real-time energy supply and demand information is the current actual energy supply and demand situation, including the total amount of energy that the virtual power plant can currently provide, that is, the real-time total energy supply, the total amount of energy currently required by each node, that is, the real-time total energy demand, and the energy status that the energy storage device can currently store or release, that is, the real-time status data. The result generated in this step is to obtain these real-time information.

[0121] In an embodiment of the present application, through monitoring devices such as sensors on various devices in the virtual power plant, the current power generation of the power supply equipment is collected in real time to obtain the real-time total energy supply, the current power consumption of each energy participating node is collected to obtain the real-time total energy demand, and at the same time, the current storage capacity and charge and discharge capacity of the energy storage device are collected to obtain real-time status data. These data are summarized as real-time energy supply and demand information. For example, the virtual power plant collects through sensors that the current solar energy and wind energy generate a total of 500 degrees, which is the total supply, and each node uses a total of 400 degrees of electricity, which is the total demand. The energy storage device can also store 150 degrees, which is the status data.

[0122] c2. Based on the comparison results of the real-time total energy supply and the real-time total energy demand, the real-time supply and demand status of the virtual power plant is determined. The real-time supply and demand status includes: sufficient supply, balanced supply and insufficient supply.

[0123] Among them, the real-time total energy supply is the total amount of energy that the current virtual power plant can actually provide, the real-time total energy demand is the total amount of energy actually required by each node at present, and the real-time supply and demand status is the relationship between the current supply and demand, including sufficient supply, balance or insufficient supply. The result generated in this step is to determine the real-time supply and demand status of the virtual power plant.

[0124] In an embodiment of the present application, the real-time total energy supply obtained in step c1 is compared with the real-time total energy demand. If the total supply is significantly greater than the total demand, for example, the supply is more than 20% more than the demand, it is determined that the supply is sufficient; if the two are approximately equal and the difference is within 5%, it is determined that the supply is balanced; if the total supply is significantly less than the total demand, for example, the supply is more than 20% less than the demand, it is determined that the supply is insufficient. For example, the total supply is 500 degrees and the total demand is 400 degrees, 500 minus 400 equals 100 degrees, and the supply is 25% more than the demand, so it is determined that the supply is sufficient.

[0125] c3. If the real-time supply and demand status is sufficient supply, the non-essential load demand in the global demand adjustment plan is adjusted to obtain a first adjustment rule.

[0126] Among them, the real-time supply and demand status is the supply and demand relationship determined in step c2. Sufficient supply is a state in which the total supply is significantly greater than the total demand. The global demand adjustment plan is an overall arrangement for coordinating the electricity consumption of each node. Non-essential load demand is non-essential electricity demand in the current period, such as electricity consumption for entertainment equipment. The first adjustment rule is an adjustment criterion formulated for sufficient supply. The result of this step is to obtain the first adjustment rule.

[0127] In an embodiment of the present application, when the real-time supply and demand status is sufficient supply, the electricity demand of each node in the global demand adjustment plan is analyzed, and necessary loads such as residents' basic living electricity consumption and non-essential loads such as shopping malls' decorative electricity consumption during non-business hours are distinguished, and adjustment rules are formulated: appropriately postpone or reduce the electricity consumption time and power of non-essential loads to avoid energy waste. For example, when the supply is sufficient, it is planned to postpone the electricity consumption of the shopping mall's decorative lights from the current period to the low electricity consumption period, forming the first adjustment rule.

[0128] c4. Alternatively, if the real-time supply and demand status is insufficient, the non-critical power generation output in the power dispatch plan is adjusted to obtain a second adjustment rule.

[0129] Among them, the real-time supply and demand status is the relationship determined in step c2, insufficient supply is a state in which the total supply is significantly less than the total demand, the power dispatch plan is the arrangement guiding power distribution, non-critical power generation output is the power generation of power generation equipment that does not currently have to be operated in priority, such as low-efficiency generators, and the second adjustment rule is the adjustment criterion formulated for insufficient supply. The result of generating this step is to obtain the second adjustment rule.

[0130] In an embodiment of the present application, when the real-time supply and demand status is insufficient supply, the power generation arrangements of each power supply equipment in the power dispatch plan are analyzed, and critical power generation outputs such as high-efficiency wind power equipment and non-critical power generation outputs such as low-efficiency diesel generators are distinguished. Adjustment rules are formulated: increase the power generation of critical power generation outputs, reduce or suspend non-critical power generation outputs, and give priority to ensuring the power consumption of necessary loads. For example, when the supply is insufficient, it is planned to let high-efficiency wind farms generate more electricity and suspend low-efficiency generators, forming a second adjustment rule.

[0131] c5. Alternatively, if the real-time supply and demand status is in supply balance, a third adjustment rule is generated based on the real-time status data of the energy storage device.

[0132] The real-time supply and demand status is the relationship determined in step c2, supply balance is a state in which supply and demand are approximately equal, the real-time status data of the energy storage device is its current storage and charging and discharging capacity, and the third adjustment rule is an adjustment criterion formulated for supply balance. The result of this step is the third adjustment rule.

[0133] In an embodiment of the present application, when the real-time supply and demand status is insufficient supply, the power generation arrangements of each power supply equipment in the power dispatch plan are analyzed, and critical power generation outputs such as high-efficiency wind power equipment and non-critical power generation outputs such as low-efficiency diesel generators are distinguished. Adjustment rules are formulated: increase the power generation of critical power generation outputs, reduce or suspend non-critical power generation outputs, and give priority to ensuring the power consumption of necessary loads. For example, when the supply is insufficient, it is planned to let high-efficiency wind farms generate more electricity and suspend low-efficiency generators, forming a second adjustment rule.

[0134] This application provides the following specific example: Virtual Power Plant A obtains real-time energy supply and demand information through monitoring devices on each device: 200 kWh of solar power and 300 kWh of wind power, resulting in a total real-time energy supply of 200 + 300 = 500 kWh. Residential electricity consumption is 100 kWh, factories 250 kWh, and shopping malls 50 kWh, resulting in a total real-time energy demand of 100 + 250 + 50 = 400 kWh. The energy storage device currently has 70 kWh of storage capacity. A comparison reveals that supply exceeds demand by 100 kWh, determining that the real-time supply and demand status is sufficient. An analysis of the global demand adjustment plan identifies residential lighting as essential load and shopping mall billboards as non-essential. A first adjustment rule is established: reduce the current power consumption of the shopping mall billboards by half and delay their resumption until the evening. If the subsequent supply is insufficient, for example, supply is 300 kWh and demand is 400 kWh, a second adjustment rule is established: increase wind power generation and suspend diesel generators. If the supply is balanced and there is storage capacity, a third adjustment rule is established: store 10% of the current generated power.

[0135] By executing steps c1 to c5, the embodiment of the present application obtains real-time energy supply and demand information to provide an accurate basis for determining the current supply and demand relationship. After determining the real-time supply and demand status, adjustment rules are formulated for sufficient, insufficient, or balanced supply, allowing for flexible response to different situations. When the supply is sufficient, non-essential loads are adjusted to avoid waste. When the supply is insufficient, power generation output is optimized to ensure necessary demand. When the supply is balanced, energy storage is used to maintain stability. This improves the rationality of energy utilization as a whole, enhances the ability of the virtual power plant to cope with supply and demand fluctuations, and ensures more accurate energy supply and demand matching and more stable operation.

[0136] Figure 2 A schematic diagram of the structure of an AI-based virtual power plant energy supply and demand precise matching system provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire the output data sequence of the power equipment in the virtual power plant and the electricity price data sequence of the power market.

[0137] The adjustment module 22 is used to dynamically adjust the initial charge and discharge power of the energy storage equipment in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain the target charge and discharge power.

[0138] The optimization module 23 is used to integrate and optimize the load forecasting models of various energy participating nodes in the virtual power plant based on the target charging and discharging power using a federated learning method to generate a global demand regulation plan.

[0139] The generation module 24 is used to generate a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model.

[0140] The matching module 25 is used to coordinate the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant to achieve accurate matching of energy supply and demand of the virtual power plant based on AI.

[0141] Figure 2 The AI-based virtual power plant energy supply and demand precise matching system can perform Figure 1 The implementation principle and technical effects of the AI-based virtual power plant energy supply and demand precise matching method described in the illustrated embodiment will not be repeated here. The specific manner in which each module and unit performs operations in the above-mentioned AI-based virtual power plant energy supply and demand precise matching system has been described in detail in the embodiments of the method and will not be elaborated here.

[0142] In one possible design, Figure 2 The AI-based virtual power plant energy supply and demand precise matching system of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0143] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0144] The processing component 32 is used to execute the following process: collecting the output data sequence of the power equipment in the virtual power plant and the electricity price data sequence of the power market; dynamically adjusting the initial charge and discharge power of the energy storage equipment in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain the target charge and discharge power; based on the target charge and discharge power, using the federated learning method to integrate and optimize the load forecasting model of each energy participating node in the virtual power plant to generate a global demand adjustment plan; based on the output data sequence and the electricity price data sequence, using a pre-trained scheduling model to generate a power dispatching plan; and coordinating the power dispatching plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant to achieve accurate matching of energy supply and demand of the virtual power plant based on AI.

[0145] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0146] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0147] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0148] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0149] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0150] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0151] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is an AI-based method for accurately matching energy supply and demand in a virtual power plant.

[0152] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0153] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0154] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An AI-based method for accurately matching energy supply and demand in virtual power plants, characterized in that: include: Collect output data series of power equipment in virtual power plants and electricity price data series in the power market; Dynamically adjusting the initial charge and discharge power of the energy storage device in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain a target charge and discharge power; Based on the target charge and discharge power, a federated learning approach is used to integrate and optimize the load forecasting models of the energy participating nodes in the virtual power plant to generate a global demand regulation plan; Based on the output data sequence and the electricity price data sequence, generating a power dispatch plan using a pre-trained dispatch model; The power dispatching plan and the global demand regulation plan are coordinated in the energy supply and demand matching process of the virtual power plant to achieve accurate energy supply and demand matching of the virtual power plant based on AI.

2. The AI-based virtual power plant energy supply and demand precise matching method according to claim 1 is characterized in that: The generating of a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model includes: Inputting the output data sequence and the electricity price data sequence into a pre-trained scheduling model, and dividing the output data sequence into a plurality of data segments according to a preset time length by a division unit in the scheduling model; Determining the output value fluctuation range of each data segment by a determination unit in the scheduling model, and obtaining the output fluctuation characteristics corresponding to each data segment based on the output value fluctuation range; The calculation unit in the scheduling model calculates the electricity price data sequence to obtain a price change amplitude value, and generates a price fluctuation trend based on the price change amplitude value; The processing unit in the dispatching model associates the output fluctuation characteristics with the price fluctuation trend to obtain a correlation between the output fluctuation characteristics and the price fluctuation trend, and generates a power dispatching plan based on the correlation.

3. The AI-based virtual power plant energy supply and demand precise matching method according to claim 2 is characterized in that: The step of calculating the electricity price data sequence to obtain a price change amplitude value, and generating a price fluctuation trend based on the price change amplitude value, includes: Dividing the electricity price data sequence into a plurality of electricity price data intervals according to the preset time length; Comparing the electricity price data corresponding to adjacent time points within each of the electricity price data intervals to obtain a price change amplitude value within each of the electricity price data intervals; Determining the price fluctuation direction of each electricity price data interval based on the price change amplitude value of each electricity price data interval; The price change magnitude values ​​and the price fluctuation directions of all the electricity price data intervals are integrated in chronological order to generate a price fluctuation trend.

4. The AI-based virtual power plant energy supply and demand precise matching method according to claim 1 is characterized in that: Based on the target charge and discharge power, the load forecasting model of each energy participating node in the virtual power plant is integrated and optimized by using a federated learning method to generate a global demand adjustment plan, including: Based on the target charge and discharge power, adjusting the optimization parameters of the load forecasting model; By means of each energy participating node in the virtual power plant, according to the adjusted optimization parameters, the calculation results of the corresponding load forecasting model are updated, and based on the updated calculation results, the key parameters of the load forecasting model are extracted; The key parameters are sent to the central coordination node of the virtual power plant through each energy participating node. Based on the federated learning method, the key parameters of all load forecasting models are comprehensively adjusted by the central coordination node to generate final unified parameters applicable to the whole world. Sending the unified parameter to each energy participating node, so that each energy participating node optimizes the corresponding load forecasting model according to the unified parameter; Based on the central coordination node, the output results of all optimized load forecasting models are integrated to generate a global demand regulation plan.

5. The AI-based virtual power plant energy supply and demand precise matching method according to claim 4 is characterized in that: The key parameters are sent to the central coordination node of the virtual power plant through each energy participating node. Based on the federated learning method, the key parameters of all load forecasting models are comprehensively adjusted by the central coordination node to generate final unified parameters applicable to the global load forecasting model, including: Based on the federated learning method, the central coordination node is used to determine whether the difference value of each key parameter exceeds a preset range, and the key parameter whose difference value exceeds the preset range is determined as the difference parameter; Based on the central coordination node, the difference parameters are gradient-adjusted, and the difference parameters after gradient adjustment are combined with the indifference parameters to form preliminary unified parameters, where the indifference parameters are key parameters whose difference values ​​do not exceed a preset range; Distributing the preliminary unified parameters to each energy participating node, each node using a local validation data set to calculate the model performance index corresponding to the preliminary unified parameters, and feeding back the performance change between the model performance index and the original performance index to the central coordination node; The central coordination node amends the preliminary unified parameters based on the performance variation to generate final unified parameters applicable to the global situation.

6. The AI-based virtual power plant energy supply and demand precise matching method according to claim 1 is characterized in that: The process of coordinating the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant includes: Dividing the power dispatch plan and the global demand adjustment plan according to preset time lengths to obtain a plurality of time units corresponding to each; Matching the time units of the power dispatch plan with the time units of the global demand adjustment plan in chronological order, so that the plan contents of both parties in each time unit group form a time association, and obtaining a time association plan corresponding to each time unit group; Based on the time association plan corresponding to the same time unit group, identifying the inconsistent parts between the power dispatch plan and the global demand adjustment plan in the same time unit as the plan conflict points; Determining adjustment rules for the plan conflict points based on real-time energy supply and demand information of the virtual power plant; Based on the adjustment rules, the plan conflict points in each of the time-related plans are corrected to form a collaborative plan, and the collaborative plan is applied to the energy supply and demand matching process of the virtual power plant.

7. The AI-based virtual power plant energy supply and demand precise matching method according to claim 6 is characterized in that: Determining the adjustment rules for the plan conflict points based on the real-time energy supply and demand information of the virtual power plant includes: Acquire real-time energy supply and demand information of the virtual power plant, wherein the real-time energy supply and demand information includes: the real-time total energy supply and real-time total energy demand of the virtual power plant and real-time status data of energy storage equipment; Determining a real-time supply and demand state of the virtual power plant based on a comparison result of the real-time total energy supply and the real-time total energy demand, wherein the real-time supply and demand state includes: sufficient supply, balanced supply, and insufficient supply; If the real-time supply and demand status is in sufficient supply, adjusting the non-essential load demand in the global demand adjustment plan to obtain a first adjustment rule; Alternatively, if the real-time supply and demand status is in short supply, the non-critical power generation output in the power dispatch plan is adjusted to obtain a second adjustment rule; Alternatively, if the real-time supply and demand status is in supply balance, a third adjustment rule is generated based on the real-time status data of the energy storage device.

8. An AI-based virtual power plant energy supply and demand precise matching system, characterized by: include: The acquisition module is used to collect the output data series of the power equipment in the virtual power plant and the electricity price data series of the power market; an adjustment module, configured to dynamically adjust the initial charge and discharge power of the energy storage device in the virtual power plant according to the output data sequence and the electricity price data sequence to obtain a target charge and discharge power; an optimization module for fusing and optimizing the load forecasting models of the energy participating nodes in the virtual power plant based on the target charge and discharge power using a federated learning approach to generate a global demand regulation plan; A generation module, configured to generate a power dispatch plan based on the output data sequence and the electricity price data sequence using a pre-trained dispatch model; A matching module is used to coordinate the power dispatch plan and the global demand adjustment plan in the energy supply and demand matching process of the virtual power plant to achieve accurate matching of energy supply and demand of the virtual power plant based on AI.

Images