A resource-optimized virtual power plant control method and device

Abstract

The application discloses a resource-optimized virtual power plant control method and device, and relates to the technical field of virtual power plant control. The method comprises the following steps: collecting multi-dimensional monitoring data of a virtual power plant, inputting the virtual power plant state prediction model, performing state prediction, and obtaining a virtual power plant state prediction result; inputting the virtual power plant state prediction result into a virtual power plant resource allocation optimization model, performing resource allocation scheme generation, and obtaining a virtual power plant resource allocation scheme; inputting the virtual power plant resource allocation scheme into a virtual power plant digital twin model, performing digital twin simulation verification, and obtaining a corrected virtual power plant resource allocation scheme; inputting the corrected virtual power plant resource allocation scheme into a virtual power plant cooperative control model, performing control instruction generation, obtaining a virtual power plant control instruction, and delivering the virtual power plant control instruction to an execution terminal. The method solves the problems of single data dimension, insufficient prediction accuracy, extensive resource allocation, lack of closed-loop verification and complex cooperative control in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual power plant control, and particularly relates to a resource-optimized virtual power plant control method and device. BACKGROUND

[0002] With the large-scale access of renewable energy, virtual power plant, as an innovative mode of aggregating distributed energy, controllable load and energy storage resources, participating in power market transactions and grid auxiliary services, is attracting more and more attention. The core of virtual power plant lies in the optimal scheduling and control of internal resources to maximize economic benefits, stable operation of the grid or to meet user demand.

[0003] However, the existing virtual power plant control method faces many challenges:

[0004] 1) Single data dimension: Traditional control methods often only focus on part of the key data, such as load and power generation, ignoring multi-dimensional information such as grid state, user behavior, environmental factors, etc., resulting in limited prediction accuracy and control effect.

[0005] 2) Insufficient prediction accuracy: The grid load, renewable energy output (such as photovoltaic and wind power) and user adjustable load have strong uncertainty and volatility, and existing prediction models are difficult to accurately capture their dynamic changes, especially lacking in-depth mining of causal relationships, resulting in poor robustness of control strategies.

[0006] 3) Rough resource allocation: Resource allocation strategies are often based on simple optimization algorithms, making it difficult to balance long-term strategic planning and short-term flexible adjustment, and failing to fully utilize the temporal and spatial complementarity of internal resources of virtual power plant.

[0007] 4) Lack of closed-loop verification: The generated control scheme is often directly issued for execution, lacking simulation verification in a virtual environment, which may cause problems in actual execution, affecting system stability and control effect.

[0008] 5) Complex collaborative control: The internal resources of virtual power plant are diverse in type and characteristics, and traditional centralized or decentralized control methods are difficult to achieve efficient and flexible collaborative control, especially in multi-objective optimization and distributed decision-making. SUMMARY

[0009] The present application provides a resource-optimized virtual power plant control method and device, which solves the problems of single data dimension, insufficient prediction accuracy, rough resource allocation, lack of closed-loop verification and complex collaborative control in the prior art.

[0010] In a first aspect, the present application embodiment provides a resource-optimized virtual power plant control method, the method comprising:

[0011] The multi-dimensional monitoring data of the virtual power plant is collected, and input into a virtual power plant state prediction model to perform state prediction, to obtain a virtual power plant state prediction result;

[0012] The virtual power plant state prediction result is input into a virtual power plant resource allocation optimization model to perform resource allocation scheme generation, to obtain a virtual power plant resource allocation scheme;

[0013] The virtual power plant resource allocation scheme is input into a virtual power plant digital twin model to perform digital twin simulation verification, to obtain a corrected virtual power plant resource allocation scheme;

[0014] The corrected virtual power plant resource allocation scheme is input into a virtual power plant collaborative control model to perform control instruction generation, to obtain virtual power plant control instructions, and the control instructions are issued to corresponding execution terminals.

[0015] The technical scheme provided by the embodiments of the present application at least brings the following beneficial effects:

[0016] By collecting multi-dimensional data such as grid side, user side, environment and resource state, the operation environment and internal state of the virtual power plant can be more comprehensively understood, and compared with the traditional method which only relies on a few types of data, the accuracy and reliability of the prediction result are significantly improved; based on the prediction of the future state, the virtual power plant resource allocation optimization model can formulate a forward-looking resource allocation strategy, rather than just responding to the current demand, so as to better balance the cost, benefit and reliability, consider the characteristics of various resources comprehensively, maximize the use of all available resources under the premise of meeting the target, improve the overall energy utilization efficiency, in the face of the situation that the internal resources of the virtual power plant are of many types and have different characteristics, the optimization model can handle complex constraint conditions and generate a more optimal resource allocation scheme; the digital twin model is an accurate mapping of the actual virtual power plant, which can simulate the actual effect of the execution of the resource allocation scheme in the virtual environment, which can effectively identify and avoid potential physical conflicts, equipment over-limit or coordination failure in the scheme, and ensure the feasibility of the scheme in actual execution, through simulation verification, the performance of the scheme under different disturbances can be evaluated and corrected accordingly, thereby improving the robustness and anti-interference ability of the whole control system, testing and optimizing the scheme in the virtual environment can avoid the economic loss and operation risk caused by repeated trial and error in the actual system; the virtual power plant collaborative control model can effectively coordinate the resources scattered in different geographical locations and of different types, so that they operate like a whole, improve the control efficiency and response speed, convert the scheme after prediction, optimization and verification into specific and executable control instructions, and accurately issue them to the corresponding execution terminal, to ensure that the resource allocation scheme can be accurately and timely executed, through fine collaborative control, the virtual power plant can better achieve the goals of participating in grid regulation, peak shaving and valley filling, demand response, etc., and improve the overall operation performance and market competitiveness of the virtual power plant.

[0017] In an optional implementation, the multi-dimensional monitoring data includes grid-side monitoring data, user-side monitoring data, environmental monitoring data and resource state monitoring data.

[0018] In an optional implementation, the virtual power plant state prediction model is constructed based on an STGNN-LSTM-FCI-MLP-PC algorithm, and the virtual power plant state prediction model includes a space-time feature extraction layer constructed based on an STGNN algorithm, a time series modeling layer constructed based on an LSTM algorithm, a causal inference layer constructed based on an FCI algorithm, a virtual power plant state prediction layer constructed based on an MLP algorithm, and a physical constraint layer constructed based on a PC algorithm, which are sequentially connected.

[0019] The virtual power plant resource allocation optimization model is constructed based on an HRL algorithm, and the virtual power plant resource allocation optimization model includes a strategy layer and a tactic layer which are sequentially connected, the strategy layer is provided with a strategy agent constructed based on an MILP algorithm, and the tactic layer is provided with a tactic agent constructed based on an MPC algorithm.

[0020] The virtual power plant digital twin model includes a physical layer, a data layer, a behavior layer and a control layer.

[0021] The virtual power plant collaborative control model is based on an MARL algorithm, and the virtual power plant collaborative control model includes a centralized layer and a decentralized layer, the centralized layer is provided with a master agent constructed based on a PPO algorithm, and the decentralized layer is provided with a plurality of sub-agents constructed based on a DQN algorithm.

[0022] In an optional implementation, multi-dimensional monitoring data of the virtual power plant is collected, and is input into the virtual power plant state prediction model to perform state prediction, to obtain a virtual power plant state prediction result, including:

[0023] The multi-dimensional monitoring data of the virtual power plant is collected, and the multi-dimensional monitoring data is preprocessed to obtain preprocessed multi-dimensional monitoring data, and a corresponding space-time graph is constructed;

[0024] The space-time feature extraction layer of the virtual power plant state prediction model is used to extract space-time features of the space-time graph;

[0025] The time series modeling layer of the virtual power plant state prediction model is used to generate a time series of the space-time features;

[0026] The causal inference layer of the virtual power plant state prediction model is used to generate a causal inference result of the time series;

[0027] According to the time series and the causal inference result, the virtual power plant state prediction layer of the virtual power plant state prediction model is used to perform state prediction, to obtain an initial virtual power plant state prediction result;

[0028] The physical constraint layer of the virtual power plant state prediction model is used to perform physical constraint and correction on the initial virtual power plant state prediction result, so as to obtain a final virtual power plant state prediction result.

[0029] In an optional implementation, the virtual power plant state prediction result includes a power grid total load prediction value and a confidence interval thereof, a subdivided power consumption load prediction value of each type of user, a power generation prediction value of each type of renewable energy within the virtual power plant, and a potential response capability prediction value of the user-side adjustable load.

[0030] In an optional implementation, the virtual power plant state prediction result is input into a virtual power plant resource allocation optimization model to generate a resource allocation scheme, so as to obtain a virtual power plant resource allocation scheme, including:

[0031] Global objectives and constraint conditions of the virtual power plant resource allocation optimization model are set, and the virtual power plant state prediction result is input into the virtual power plant resource allocation optimization model;

[0032] According to the virtual power plant state prediction result, a strategic agent of a strategic layer of the virtual power plant resource allocation optimization model is used to perform resource optimization, so as to obtain a virtual power plant resource operation plan;

[0033] According to the virtual power plant resource operation plan, a tactical agent of a tactical layer of the virtual power plant resource allocation optimization model is used to generate a resource allocation scheme, so as to obtain a virtual power plant resource allocation scheme.

[0034] In an optional implementation, the virtual power plant resource allocation scheme is input into a virtual power plant digital twin model to perform digital twin simulation verification, so as to obtain a corrected virtual power plant resource allocation scheme, including:

[0035] According to the virtual power plant resource allocation scheme and multi-dimensional monitoring data, a digital twin technology is used to construct a virtual power plant digital twin model, and a corresponding simulation scene is set;

[0036] The virtual power plant resource allocation scheme is input into the virtual power plant digital twin model to perform dynamic simulation operation, and a key performance indicator in a dynamic simulation operation process is collected in real time;

[0037] According to the key performance indicator, digital twin simulation verification is performed to generate a corresponding simulation evaluation result;

[0038] According to the simulation evaluation result, the virtual power plant resource allocation scheme is dynamically adjusted and optimized to obtain a corrected virtual power plant resource allocation scheme.

[0039] In an optional implementation, according to the virtual power plant resource allocation scheme and multi-dimensional monitoring data, a digital twin technology is used to construct a virtual power plant digital twin model, and a corresponding simulation scene is set, including:

[0040] constructing a physical layer including a plurality of physical models of the virtual power plant according to the grid-side monitoring data in the multi-dimensional monitoring data;

[0041] constructing a data layer including a data warehouse according to the real-time collected data and the plurality of historical collected data in the multi-dimensional monitoring data;

[0042] constructing a behavior layer including behavior models of various resources of the virtual power plant according to the grid-side monitoring data and the resource state monitoring data in the multi-dimensional monitoring data;

[0043] mapping the virtual power plant resource allocation scheme to the digital twin space to construct a control layer including a plurality of controllers;

[0044] integrating the physical layer, the data layer, the behavior layer, and the control layer into a unified platform to form a virtual mapped virtual power plant digital twin model consistent with the actual virtual power plant topology and resources;

[0045] setting a simulation scenario of the virtual power plant digital twin model according to the virtual power plant state prediction result and the virtual power plant resource allocation scheme.

[0046] In an optional implementation, the corrected virtual power plant resource allocation scheme is input into the virtual power plant collaborative control model to generate control instructions, obtain virtual power plant control instructions, and is issued to the corresponding execution terminal, including:

[0047] allocating various resources of the virtual power plant to the sub-agents corresponding to the decentralized layer in the virtual power plant collaborative control model according to the resource state monitoring data in the multi-dimensional monitoring data, and allocating the control center of the virtual power plant to the master agent corresponding to the centralized layer in the virtual power plant collaborative control model;

[0048] setting corresponding state spaces, action spaces, objective functions, and global and local information perception capabilities for the master agent and the plurality of sub-agents, and establishing a communication protocol between the master agent and the plurality of sub-agents;

[0049] inputting the corrected virtual power plant resource allocation scheme into the virtual power plant collaborative control model and decomposing it into global actions of the action space of the master agent and specific actions of the action space of the sub-agents;

[0050] using a distributed optimization algorithm to perform collaborative control on the master agent and the plurality of sub-agents according to the collaborative control mechanism to obtain execution actions of the plurality of sub-agents;

[0051] generating control instructions according to the execution actions of the plurality of sub-agents to obtain virtual power plant control instructions, and issuing the virtual power plant control instructions to the execution terminal corresponding to each sub-agent.

[0052] In a second aspect, an embodiment of the present application provides a resource-optimized virtual power plant control device for implementing a virtual power plant control method. The device comprises:

[0053] a state prediction unit configured to collect multi-dimensional monitoring data of the virtual power plant, input a virtual power plant state prediction model, perform state prediction, and obtain a virtual power plant state prediction result;

[0054] a resource allocation unit configured to input the virtual power plant state prediction result into a virtual power plant resource allocation optimization model, generate a resource allocation scheme, and obtain a virtual power plant resource allocation scheme;

[0055] a scheme correction unit configured to input the virtual power plant resource allocation scheme into a virtual power plant digital twin model, perform digital twin simulation verification, and obtain a corrected virtual power plant resource allocation scheme;

[0056] a cooperative control unit configured to input the corrected virtual power plant resource allocation scheme into a virtual power plant cooperative control model, generate a control instruction, obtain a virtual power plant control instruction, and distribute the virtual power plant control instruction to a corresponding execution terminal.

[0057] In a third aspect, an embodiment of the present application provides an electronic device, which comprises:

[0058] at least one processor; and a memory communicatively connected to the at least one processor; wherein

[0059] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the first aspect of the present application.

[0060] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and the program is executed by a processor to implement the method according to the first aspect of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 FIG. 1 is a schematic diagram of an electronic device structure of a hardware running environment related to an embodiment of the present application;

[0062] Figure 2 FIG. 2 is a step flowchart of a resource-optimized virtual power plant control method according to an embodiment of the present application;

[0063] Figure 3 FIG. 3 is a functional unit schematic diagram of a resource-optimized virtual power plant control device according to an embodiment of the present application. DETAILED DESCRIPTION

[0064] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0065] The scheme of the present application will be further described below with reference to the drawings.

[0066] Referring to Figure 1 , Figure 1 The electronic device structure schematic diagram of the hardware running environment involved in the embodiment scheme of the present application.

[0067] As Figure 1 shown, the electronic device can include a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display, an input unit such as a keyboard, and can also include a standard wired interface, a wireless interface. The network interface 1004 can optionally include a standard wired interface, a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 can be a high-speed random access memory (RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a magnetic disk memory. The memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0068] Those skilled in the art can understand that Figure 1 the structure shown in the above does not constitute a limitation on the electronic device, and can include more or fewer components than the diagram, or combine certain components, or different component arrangements.

[0069] As Figure 1 shown, the memory 1005 as a storage medium can include an operating device, a data storage module, a network communication module, a user interface module, and an electronic program.

[0070] In Figure 1The network interface 1004 in the electronic device shown is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the electronic device of the present application can be arranged in the electronic device, and the electronic device calls the resource-optimized virtual power plant control device stored in the memory 1005 through the processor 1001, and executes the resource-optimized virtual power plant control method provided by the embodiments of the present application.

[0071] With reference to Figure 2 The embodiments of the present application provide a resource-optimized virtual power plant control method, which can specifically include the following steps:

[0072] S201: Collecting multi-dimensional monitoring data of the virtual power plant, inputting the virtual power plant state prediction model, performing state prediction, and obtaining a virtual power plant state prediction result;

[0073] S202: Inputting the virtual power plant state prediction result into a virtual power plant resource allocation optimization model, performing resource allocation scheme generation, and obtaining a virtual power plant resource allocation scheme;

[0074] S203: Inputting the virtual power plant resource allocation scheme into a virtual power plant digital twin model, performing digital twin simulation verification, and obtaining a corrected virtual power plant resource allocation scheme;

[0075] S204: Inputting the corrected virtual power plant resource allocation scheme into a virtual power plant cooperative control model, performing control instruction generation, obtaining a virtual power plant control instruction, and issuing the control instruction to a corresponding execution terminal.

[0076] By collecting multi-dimensional data such as grid side, user side, environment and resource state, the operation environment and internal state of the virtual power plant can be more comprehensively understood. Compared with the traditional method which only relies on a few types of data, the accuracy and reliability of the prediction results (such as load, power generation, and adjustable resource potential) are significantly improved. Based on the prediction of the future state, the virtual power plant resource allocation optimization model can develop a forward-looking resource allocation strategy, rather than just responding to the current demand, thereby better balancing cost, benefit and reliability. It can comprehensively consider the characteristics of various resources (such as the volatility of renewable energy, the charging and discharging characteristics of energy storage, the response speed and capacity of adjustable load), maximize the use of all available resources under the premise of meeting the target (such as grid instructions, economy, and user demand), improve overall energy utilization efficiency, and face the situation of multiple types of internal resources in the virtual power plant with different characteristics. The optimization model can handle complex constraints (such as device capacity, ramp rate, scheduling rules, etc.) to generate a better resource allocation scheme. The digital twin model is an accurate mapping of the actual virtual power plant, which can simulate the actual effect of the resource allocation scheme after execution in the virtual environment. This can effectively identify and avoid potential physical conflicts, device overruns or coordination failures in the scheme, ensuring the feasibility of the scheme in actual execution. Through simulation verification, the performance of the scheme under different disturbances (such as prediction errors, device failures, and communication delays) can be evaluated and corrected accordingly, thereby improving the robustness and anti-interference ability of the entire control system. Testing and optimization of the scheme in the virtual environment avoids the economic losses and operational risks that may be caused by repeated trial and error in the actual system. The virtual power plant collaborative control model can effectively coordinate resources (such as distributed power sources, energy storage, and adjustable loads) scattered in different geographical locations and different types, making them operate as a whole, improving control efficiency and response speed. The scheme that has been predicted, optimized and verified is converted into specific and executable control instructions and accurately issued to the corresponding execution terminal (such as energy storage converters, smart meters, and load controllers), ensuring that the resource allocation scheme can be accurately and timely executed. Through fine-grained collaborative control, the virtual power plant can better achieve the goals of participating in grid regulation, peak shaving and valley filling, demand response, and improve the overall operation performance and market competitiveness of the virtual power plant.

[0077] In an optional implementation, the multi-dimensional monitoring data includes grid side monitoring data, user side monitoring data, environment monitoring data, and resource state monitoring data.

[0078] In this embodiment, the grid side monitoring data includes real-time collected grid operation data, including but not limited to: grid topology, current load level, transmission line state, substation operation parameter, day-ahead / real-time electricity price, day-ahead / real-time power generation plan, etc.

[0079] User-side monitoring data includes user-side data collected through smart meters, Internet of Things devices, etc., including but not limited to: historical power load curve, user device type (air conditioner, water heater, electric vehicle, etc.), device operating state, user behavior pattern (such as weekday / weekend power consumption difference, specific time period power consumption habit, etc.), user willingness to participate in virtual power plant, etc.;

[0080] Environmental monitoring data includes high-precision weather data, including but not limited to: 24-72 hour weather forecast (wind speed, wind direction, temperature, humidity, cloud cover, light intensity, etc.), air quality index, special weather event warning (such as typhoon, heavy rain, extreme high / low temperature, etc.), data from weather satellites, ground weather stations, professional weather service providers, etc.;

[0081] Resource state monitoring data includes state data of various resources within the virtual power plant, including but not limited to: distributed photovoltaic output, wind power output, state of charge of energy storage system (battery), charge and discharge power limit, adjustable range and response speed of adjustable load (such as industrial motor, air conditioner temperature control), electric vehicle charging pile state and charging demand, etc.

[0082] In an optional implementation, the virtual power plant state prediction model is constructed based on a Spatio-Temporal Graph Neural Network (STGNN)-Long Short-Term Memory (LSTM)-Fast Causal Inference (FCI)-Multi-Layer Perceptron (MLP)-Physical Constraints (PC) algorithm, and the virtual power plant state prediction model includes a spatio-temporal feature extraction layer constructed based on the STGNN algorithm, a time series modeling layer constructed based on the LSTM algorithm, a causal inference layer constructed based on the FCI algorithm, a virtual power plant state prediction layer constructed based on the MLP algorithm, and a physical constraint layer constructed based on the PC algorithm, which are connected in sequence.

[0083] It is worth noting that by fusing multi-dimensional data such as grid side, user side, environment and resource state, and using advanced STGNN-LSTM-FCI-MLP-PC algorithm, the virtual power plant state prediction model not only considers spatio-temporal correlation and time series characteristics, but also introduces causal inference, which can more accurately predict grid load, renewable energy output and user response capability, providing a more reliable basis for subsequent optimization.

[0084] In this embodiment, the physical constraint layer is provided with physical constraint formulas, capacity limit constraint formulas, and physical feasibility formulas, which serve as prior knowledge constraints for the deep learning model and guide feature output and model training. The physical constraint layer checks the prediction output to ensure that it meets the physical feasibility conditions. If the prediction result does not meet the physical constraints, the output of the virtual power plant state prediction layer is corrected.

[0085] The physical constraint formula is:

[0086]

[0087] In the formula, This is a physical constraint check value; For grid interaction power planning; This represents the actual power exchange between power grids. The virtual power plant state prediction layer outputs the total grid load forecast, photovoltaic power generation forecast, wind power generation forecast, and potential response capability forecast of user-side adjustable load. For power tolerance; t For time indication; Detailed electricity load forecasts for industry, commerce, and residential use;

[0088] The capacity constraint formula is:

[0089]

[0090] In the formula, This is the value for checking capacity limit constraints; For grid interaction power planning; Rated capacity for photovoltaic, wind power, and user-side adjustable loads; These are the minimum and maximum values ​​of the confidence interval for the predicted total load of the power grid;

[0091] The formula for physical feasibility is:

[0092]

[0093] In the formula, This is the value for physical feasibility checks;

[0094] In this embodiment, a sample dataset consisting of several labeled multi-dimensional monitoring data is used for training to obtain a virtual power plant state prediction model. The training objective is to minimize the total loss and update the model parameters of the virtual power plant state prediction model.

[0095] The total loss function of the virtual power plant state prediction model includes the prediction error loss function and the physical constraint loss function;

[0096] The formula is:

[0097]

[0098] In the formula, This is the total loss function; The prediction error loss function; The physical constraint loss function; Hyperparameters used to balance the weights of physical constraint terms; For other regularization terms;

[0099] Update the model parameters of the virtual power plant state prediction model using the Adam with Weight Decay (AdamW) optimizer;

[0100] The formula for the AdamW optimizer is:

[0101]

[0102] In the formula, This is a set of model parameters, which typically include weight matrices, bias terms, etc., and determine the model's output and behavior. The training dataset contains all the input data and corresponding labels used to train the model; This is a sample drawn from the training dataset; To label multi-dimensional monitoring data; For the corresponding virtual power plant status prediction labels; This is the total loss value, used to measure the difference between the model's predictions and the true labels; The expected operation form represents the operation performed on the training dataset. For all possible input-output pairs Take the average, calculate the average of the loss function over all samples. In practical applications, a small batch of samples is usually used for approximate calculation.

[0103]

[0104] In the formula, For a moment t , t -1 model parameters; The first and second moments are estimated after bias correction for the gradient; It is a very small constant used to prevent division by zero errors; This is the weight decay coefficient, used to control the size of the weights and prevent overfitting; For a moment t , t First and second moment estimates of the gradient of -1; For gradient; This is the gradient calculation function; For a moment t The total loss value corresponding to the model parameters; The attenuation rate is estimated from the first and second moments;

[0105] The virtual power plant resource allocation optimization model is constructed based on the hierarchical reinforcement learning (HRL) algorithm. The virtual power plant resource allocation optimization model includes a strategic layer and a tactical layer connected in sequence. The strategic layer is equipped with a strategic agent constructed based on the mixed-integer linear programming (MILP) algorithm, and the tactical layer is equipped with a tactical agent constructed based on the model predictive control (MPC) algorithm.

[0106] It is worth noting that by adopting a resource allocation optimization model, long-term resource planning and economic optimization are carried out at the strategic level, while short-term and dynamic resource allocation and scheduling are carried out at the tactical level. This effectively balances overall economic benefits with real-time response needs and improves resource utilization efficiency.

[0107] The virtual power plant digital twin model includes a physical layer, a data layer, a behavior layer, and a control layer;

[0108] It is worth noting that by introducing digital twin technology for simulation verification, the actual execution effect of the control scheme can be simulated in a virtual environment, key performance indicators can be evaluated, potential problems can be identified in advance and adjustments and optimizations can be made, avoiding the risks that may be brought about by direct execution on the physical system, and improving the robustness and reliability of the control scheme.

[0109] The virtual power plant collaborative control model is based on the multi-agent reinforcement learning (MARL) algorithm. The virtual power plant collaborative control model includes a centralized layer and a decentralized layer. The centralized layer is set with a master agent constructed based on the proximal policy optimization (PPO) algorithm, and the decentralized layer is set with several sub-agents constructed based on the deep Q-network (DQN) algorithm.

[0110] It is worth noting that by adopting the virtual power plant collaborative control model, through the cooperation of the centralized and decentralized layers, the complexity of the diverse and heterogeneous resources within the virtual power plant can be handled, achieving distributed and adaptive control guided by global objectives, thus improving the flexibility and adaptability of control.

[0111] In one optional implementation, multi-dimensional monitoring data of the virtual power plant is collected and input into the virtual power plant state prediction model to perform state prediction, thereby obtaining the virtual power plant state prediction results, including:

[0112] S2011: Collect multi-dimensional monitoring data of the virtual power plant, preprocess the multi-dimensional monitoring data to obtain preprocessed multi-dimensional monitoring data, and construct the corresponding spatiotemporal map;

[0113] The formula is:

[0114]

[0115] In the formula, A spatiotemporal diagram constructed based on the power grid topology; A set of nodes (such as photovoltaic generators, substations, distribution substations, user-side loads, etc.); For the set of edges (transmission and distribution lines), the node features are several related features of the multi-dimensional monitoring data (e.g., electricity consumption data, voltage, equipment status, etc.). The edge features can represent line impedance, power transmission capacity, etc.

[0116] Preprocessing includes cleaning, denoising, normalizing, and removing outliers from the collected data;

[0117] S2012: Use the spatiotemporal feature extraction layer of the virtual power plant state prediction model to extract the spatiotemporal features of the spatiotemporal map;

[0118] The formula is:

[0119]

[0120] In the formula, For a moment t , t -1's spatiotemporal characteristics; For spatiotemporal feature extraction functions; This is the node feature matrix of the spatiotemporal graph;

[0121] S2013: Time series modeling layer using virtual power plant state prediction model to generate time series of spatiotemporal features;

[0122] The formula is:

[0123]

[0124] In the formula, For a moment t , t -1 time series; Functions for modeling time series; For a moment t , t-1, ..., t - k The spatiotemporal characteristics; k The length of the time window;

[0125] S2014: Using the causal inference layer of the virtual power plant state prediction model, generate causal inference results for time series.

[0126] The formula is:

[0127]

[0128] In the formula, This is a result of causal inference; It is a causal inference function; This represents the change in the node feature matrix of the spatiotemporal graph;

[0129] S2015: Based on the time series and causal inference results, the virtual power plant state prediction layer of the virtual power plant state prediction model is used to perform state prediction and obtain the initial virtual power plant state prediction results.

[0130] The formula is:

[0131]

[0132] In the formula, For a moment t The time-to-time output of the virtual power plant state prediction model t + h The initial virtual power plant state prediction results; This is a virtual power plant state prediction function; h To predict the step size;

[0133]

[0134] S2016: Using the physical constraint layer of the virtual power plant state prediction model, the initial virtual power plant state prediction results are physically constrained and corrected to obtain the final virtual power plant state prediction results.

[0135] like If the requirements are not met, physical constraints and corrections are necessary.

[0136] The formula is:

[0137]

[0138] In the formula, Correction values ​​for photovoltaic, wind power, and user-side adjustable loads; The function to update the virtual power plant state prediction vector; This is the corrected virtual power plant state prediction vector, which is the final virtual power plant state prediction result.

[0139] In this embodiment, the virtual power plant state prediction results include future... h The forecast values ​​of total grid load at any given time (e.g., the next 24 hours, with a time granularity of 15 minutes) and their confidence intervals, the detailed electricity load forecast values ​​for various types of users (industrial, commercial, and residential), the power generation forecast values ​​for various renewable energy sources (photovoltaic and wind power) within the virtual power plant, and the potential response capability forecast values ​​for user-side adjustable loads.

[0140] In one optional implementation, the virtual power plant state prediction results are input into the virtual power plant resource allocation optimization model to generate a resource allocation scheme. This results in a virtual power plant resource allocation scheme that comprehensively considers grid demand, resource characteristics, user comfort, and multiple objectives, including:

[0141] S2021: Set the global objective and constraints of the virtual power plant resource allocation optimization model, and input the virtual power plant state prediction results into the virtual power plant resource allocation optimization model;

[0142] The formula is:

[0143]

[0144] In the formula, To minimize the optimization objective; The objective function is the global objective function. The total operating cost of the virtual power plant; This represents the total carbon emissions of the virtual power plant. The deviation between the actual output of the virtual power plant and the net load commanded by the grid or predicted reflects the plant's ability to support the grid. This is a penalty for violating constraints, such as equipment exceeding limits or excessive response deviation; The global target weight;

[0145] S2022: Based on the virtual power plant status prediction results, the strategic agent of the strategic layer of the virtual power plant resource allocation optimization model is used to optimize resources and obtain the virtual power plant resource operation plan.

[0146] It is worth noting the relevant indicators of strategic intelligent agents:

[0147] Time scale: Days in advance or longer periods (e.g., weeks);

[0148] Optimization objectives: The main objectives are to maximize economic benefits (considering electricity purchase costs, electricity sales revenue, and ancillary service revenue) and carbon emission reduction benefits, while also taking into account grid stability (such as reducing peak-valley differences and minimizing the impact on the grid).

[0149] The formula is:

[0150]

[0151] In the formula, Optimize the objective function for the strategic level; To predict the total operating cost of a long-term virtual power plant; To predict the total carbon emissions of a long-term virtual power plant; Optimize target weights for the strategic level;

[0152] Constraints include, but are not limited to: grid load balance constraints, physical characteristic constraints of various resources (power generation, energy storage, load) (such as charging and discharging power limits, upper and lower limits of state of charge (SOC), and ramp rate), grid security constraints (such as line power flow limits), constraints to ensure basic electricity demand of users, and constraints to user comfort (such as allowable temperature control range variations).

[0153] Optimization method: Mixed integer linear programming (MILP) is used to solve the multi-objective optimization problem, and the weight coefficient method or ε-constraint method is introduced to handle the multi-objective problem;

[0154] Input: Virtual power plant state prediction results;

[0155] Output: Using the virtual power plant status prediction results, formulate a preliminary virtual power plant resource operation plan, including energy storage charging and discharging plan, controllable load start-up / stop / regulation plan, and power purchase and sale plan with the main grid.

[0156] S2023: Based on the virtual power plant resource operation plan, the tactical agent of the tactical layer of the virtual power plant resource allocation optimization model generates a resource allocation scheme to obtain the virtual power plant resource allocation scheme.

[0157] It is worth noting the relevant metrics for tactical agents:

[0158] Time scale: real-time or intraday, with finer time granularity (e.g., 5-15 minutes).

[0159] Optimization objective: Based on the virtual power plant resource operation plan, make dynamic adjustments according to the latest real-time data and forecasts to cope with uncertainties and maximize short-term operational efficiency and response speed;

[0160] The formula is:

[0161]

[0162] In the formula, Optimize the objective function for the tactical layer; The total operating cost of the virtual power plant in the operation plan; The total carbon emissions of the virtual power plant in the operation plan; The deviation between the actual output of the virtual power plant in the operation plan and the grid command or predicted net load; These are penalties for violating constraints in the operational plan, such as equipment exceeding limits or excessive response deviations. Optimize target weights at the tactical level;

[0163] Optimization method: Model predictive control (MPC) is adopted to establish a more refined dynamic model, taking into account changes in equipment status, short-term fluctuations in user behavior, etc. The optimization decision variables include: the energy storage charging and discharging power at the current moment and several future time periods, the real-time adjustment command of controllable load, and the real-time power exchange plan with the main grid. The constraints are more stringent, and the equipment status and grid constraints need to be verified in real time.

[0164] Inputs: Virtual power plant resource operation plan issued by the strategic level, real-time collected grid-side monitoring data and resource status monitoring data, and virtual power plant status prediction results;

[0165] Output: Virtual power plant resource allocation scheme, that is, the precise resource allocation decision of each resource of the virtual power plant (including the adjustable load of a single user) (such as energy storage power resource allocation decision, start-up / shutdown / power adjustment decision of a single adjustable device).

[0166] In this implementation, the strategic layer converts the optimization objectives (such as the daily charging and discharging target of energy storage and the start-stop window of controllable load) corresponding to the virtual power plant status prediction results into plans and issues them to the tactical layer. The tactical layer feeds back the real-time operating status and the constraints and conflicts encountered (such as equipment failure and changes in grid commands) to the strategic layer. The strategic layer can adjust the plan for the next day accordingly. The two layers exchange information through a preset communication interface and protocol to ensure the consistency and adaptability of the virtual power plant resource allocation scheme.

[0167] In one optional implementation, the virtual power plant resource allocation scheme is input into the virtual power plant digital twin model, and digital twin simulation verification is performed to obtain a corrected virtual power plant resource allocation scheme, including:

[0168] S2031: Based on the virtual power plant resource allocation scheme and multi-dimensional monitoring data, use digital twin technology to construct a digital twin model of the virtual power plant and set up corresponding simulation scenarios;

[0169] S2032: Input the virtual power plant resource allocation scheme into the virtual power plant digital twin model, perform dynamic simulation, and collect key performance indicators in real time during the dynamic simulation process;

[0170] It is worth noting that key performance indicators include, but are not limited to: load tracking error, grid command response speed and accuracy, resource utilization (energy storage, controllable load), user comfort indicators (such as indoor temperature fluctuation range), system operating costs, carbon emissions, and grid safety indicators (such as voltage and frequency stability, and line power flow exceeding limits).

[0171] S2033: Based on key performance indicators, perform digital twin simulation verification and generate corresponding simulation evaluation results;

[0172] It is worth noting that the simulation evaluation results are used to assess the performance of the virtual power plant resource allocation scheme in the simulation environment and identify potential problems, such as the scheme may lead to equipment overload, voltage exceeding limits, frequency exceeding limits, decreased user comfort, and power grid safety risks.

[0173] Simulation evaluation methods include:

[0174] Performance evaluation: Calculate key indicators in the simulation results, such as net output deviation, equipment utilization rate, cost, and emissions, and compare them with the optimization target;

[0175] Robustness assessment: Introduce disturbance scenarios such as prediction error, equipment failure, and user response deviation to evaluate the anti-interference capability of the control scheme;

[0176] Safety assessment: Check for equipment overload, voltage exceeding limits, frequency exceeding limits, decreased user comfort, and power grid safety risks, etc.

[0177] S2034: Based on the simulation evaluation results, the virtual power plant resource allocation scheme is dynamically adjusted and optimized to obtain the revised virtual power plant resource allocation scheme;

[0178] It is worth noting that if the simulation evaluation result is "if the verification passes", the virtual power plant resource allocation scheme will be directly output as the corrected virtual power plant resource allocation scheme.

[0179] If the simulation evaluation result is that the verification fails (e.g., a safety issue or excessive deviation occurs): relevant information is fed back to step S202, and the resource allocation scheme is regenerated (adjusting the target weight, constraints, or predicted scenario), or the virtual power plant resource allocation scheme is directly modified (e.g., reducing power commands or adjusting charging and discharging strategies). The modified virtual power plant resource allocation scheme is then imported back into the digital twin model for simulation verification until the scheme meets the preset performance indicators and safety requirements.

[0180] In one alternative implementation, based on the virtual power plant resource allocation scheme and multi-dimensional monitoring data, a digital twin model of the virtual power plant is constructed using digital twin technology, and a corresponding simulation scenario is set, including:

[0181] S20311: Based on the actual geographical information of the virtual power plant, the power grid topology, and equipment parameters (generators, transformers, lines, energy storage, loads, etc.) corresponding to the grid-side monitoring data in the multi-dimensional monitoring data, establish a high-fidelity physical model (such as a synchronous generator model, an energy storage system model (including battery aging and efficiency curves), and a load model), and construct a physical layer that includes several physical models of the virtual power plant.

[0182] S20312: Based on the real-time data collected from the multi-dimensional monitoring data and several historical data collected, construct a data layer including a data warehouse. The data warehouse is used for model initialization and state updates.

[0183] S20313: Based on the grid-side monitoring data and resource status monitoring data in the multi-dimensional monitoring data, construct a behavior layer including the behavior models of various resources of the virtual power plant (such as the dynamic response model of the generator, the charging and discharging efficiency model of the energy storage system, and the dynamic response model of the load (considering user comfort)).

[0184] S20314: Map the virtual power plant resource allocation scheme to the digital twin space and construct a control layer including several controllers;

[0185] S20315: Integrate the physical layer, data layer, behavior layer, and control layer into a unified platform to form a virtual power plant digital twin model that is consistent with the topology and resources of the actual virtual power plant.

[0186] S20316: Based on the virtual power plant status prediction results and the virtual power plant resource allocation scheme, set the simulation scenario of the virtual power plant digital twin model.

[0187] In this embodiment, based on the virtual power plant status prediction results and the virtual power plant resource allocation scheme, the simulation time range and step size of the simulation scenario are set. The virtual power plant status prediction results are used as the simulation input of the virtual power plant digital twin model. Various typical and extreme scenarios can be set for testing, such as: normal operation scenario under accurate prediction, response scenario when prediction error is large, equipment sudden failure scenario, emergency change scenario of power grid command, and extreme weather impact scenario.

[0188] In one optional implementation, the modified virtual power plant resource allocation scheme is input into the virtual power plant collaborative control model to generate control commands, thereby obtaining virtual power plant control commands, which are then sent to the corresponding execution terminals, including:

[0189] S2041: Based on the resource status monitoring data in the multi-dimensional monitoring data, the various resources of the virtual power plant (large generator sets, distributed photovoltaics, energy storage systems, adjustable load clusters, and single important adjustable loads) are allocated to the sub-intelligent agents corresponding to the decentralized layer in the virtual power plant collaborative control model. Each sub-intelligent agent has a local model, status information, and limited decision-making capabilities. The control center of the virtual power plant is allocated to the master intelligent agent corresponding to the centralized layer in the virtual power plant collaborative control model.

[0190] For example, the total energy storage charging and discharging power decision is assigned to the sub-agents of specific battery cells based on the SOC, state of health (SOH) and safety limits of each battery cell, while the global decision is assigned to the master agent of the control center of the virtual power plant, such as the total energy storage charging and discharging power and the total adjustable load adjustment amount.

[0191] S2042: Set up the corresponding state space, action space, objective function, and global and local information perception capabilities for the main intelligent agent and several sub-intelligent agents, and establish a communication protocol between the main intelligent agent and several sub-intelligent agents.

[0192] It is worth noting that the state space of a sub-agent typically includes the state of local resources, such as SOC, current power, device status, security limits, etc., and may also include local information received from the main agent or other sub-agents; the state space of a sub-agent includes specific operations defined on the resources it is responsible for, such as the charging and discharging power range of a specific battery cell, start and stop commands for a single adjustable load, power adjustment range, start and stop time windows, etc.

[0193] The state space of the main agent typically includes global state information, such as the total power balance demand of the virtual power plant, electricity price signals, grid constraints, and the aggregated state (or key state) of each sub-agent; the state space of the main agent includes decisions that define the global level, such as the total energy storage charging and discharging power command range, the total adjustable load adjustment range, and resource scheduling priority commands.

[0194] The objective function of a sub-agent is usually local, such as maximizing its own efficiency, minimizing operating costs, maximizing response benefits, and satisfying its own security constraints, while also needing to respond to the global decisions of the master agent; the objective function of the master agent is global, such as minimizing the overall operating cost of the virtual power plant, maximizing the overall benefits of the virtual power plant (e.g., participating in the electricity market), satisfying grid dispatch instructions, maintaining the internal power balance of the virtual power plant, and optimizing resource allocation.

[0195] Information perception capability setting: Define the scope and methods (global information, local information, neighbor information) that the main intelligent agent and sub-intelligent agents can acquire.

[0196] Communication protocol establishment: Define how the main intelligent agent and sub-intelligent agents, and how sub-intelligent agents exchange information (such as issuing instructions, reporting status, coordinating requests, conflict resolution information, etc.).

[0197] S2043: Input the revised virtual power plant resource allocation scheme into the virtual power plant collaborative control model, and decompose it into the global actions of the main agent's action space and the specific actions of the sub-agents' action space.

[0198] It is worth noting that taking high-level decisions in the resource allocation scheme (such as the total energy storage charging and discharging power demand and the total adjustable load adjustment amount determined based on the optimization results) as the global action input of the main intelligent agent, and decomposing the global action of the main intelligent agent into specific tasks or expected outputs that need to be executed by the sub-intelligent agents, provides a target for collaborative control.

[0199] S2044: Based on the collaborative control mechanism, a distributed optimization algorithm is used to coordinate the control of the main agent and several sub-agents to obtain the execution actions of the sub-agents.

[0200] It is worth noting that distributed optimization algorithms, such as multi-agent cooperative control based on game theory, model the overall optimization problem of the virtual power plant as a multi-agent game problem. Each sub-agent adjusts its own behavior through iterative calculation based on local information and neighbor information in order to achieve the global optimization goal.

[0201] Inputs: the current state of the virtual power plant, the global actions / goals of the main agent, the local states of the sub-agents, and historical interaction information;

[0202] Cooperative control: A repeated game theory algorithm is adopted: the master agent acts as the leader, which publishes global goals and constraints; each agent acts as a follower, which adjusts its own output to respond to the leader's decision while satisfying its own constraints and local goals, and coordinates its actions through local communication to avoid conflicts (such as multiple loads adjusting significantly at the same time, causing power grid fluctuations).

[0203] Leader (main agent) actions: Based on global goals (such as resource allocation schemes) and aggregated information, issue global instructions or expectations (such as "total energy storage needs to provide XX kW of discharge power", "load cluster needs to reduce XX kW");

[0204] Follower (sub-agent) actions:

[0205] Each sub-agent (such as energy storage sub-agent A and load sub-agent B) receives global instructions and its own state.

[0206] Under the premise of satisfying its own constraints (such as SOC range, power limit, equipment characteristics) and local objectives (such as the energy storage sub-agent may want high efficiency, and the load sub-agent may want fast response speed), calculate how it should act (such as energy storage A decides to discharge Y kW, and load B decides to reduce ZZ kW).

[0207] Sub-agents coordinate through local communication to avoid conflicts (such as simultaneous large-scale adjustment of multiple loads).

[0208] The sub-agent feeds back its own decision (or adjusted decision) to the main agent (optional) or executes it directly;

[0209] The content of the collaborative control mechanism is as follows: When a sub-agent generates conflicting actions based on global instructions or local optimization (such as receiving charging and discharging instructions at the same time), the conflict is resolved according to preset priorities (such as energy storage taking priority over load, or based on resource importance) or negotiation mechanisms (communication and coordination between agents).

[0210] When a sub-agent (and the resources it is responsible for) fails, the master agent or the remaining sub-agents need to quickly detect the failure and reallocate the tasks that the sub-agent was responsible for to other available resources or agents, and adjust global and local decisions.

[0211] Output: The specific actions generated by each sub-agent and optimized through collaboration (e.g., energy storage unit A discharges YkW, load unit B reduces ZZkW).

[0212] S2045: Based on the actions of several sub-intelligent agents, control instructions are generated to obtain virtual power plant control instructions, and the virtual power plant control instructions are sent to the execution terminal corresponding to each sub-intelligent agent.

[0213] For example, specific charging and discharging power control commands for the energy storage system (such as "Energy storage converter 1, set the discharge power to Y kW"), SOC management commands, etc., can be generated. Equipment start-up and shutdown commands for adjustable loads can be generated (such as "Load controller A, start equipment X"), power adjustment commands (such as "Load controller B, reduce equipment Y power to ZZ kW"), and timing control commands (such as "Load controller C, maintain equipment Z power at W kW before time T").

[0214] This invention also provides a resource-optimized virtual power plant control device, referring to... Figure 3 The diagram shows a functional unit diagram of a resource-optimized virtual power plant control device 300 according to the present invention. The device may include the following units:

[0215] The state prediction unit 301 is used to collect multi-dimensional monitoring data of the virtual power plant, input it into the virtual power plant state prediction model, perform state prediction, and obtain the virtual power plant state prediction result.

[0216] The resource allocation unit 302 is used to input the virtual power plant state prediction results into the virtual power plant resource allocation optimization model, generate a resource allocation scheme, and obtain the virtual power plant resource allocation scheme.

[0217] The scheme correction unit 303 is used to input the virtual power plant resource allocation scheme into the virtual power plant digital twin model, perform digital twin simulation verification, and obtain the corrected virtual power plant resource allocation scheme.

[0218] The collaborative control unit 304 is used to input the modified virtual power plant resource allocation scheme into the virtual power plant collaborative control model, generate control commands, obtain virtual power plant control commands, and send them to the corresponding execution terminals.

[0219] Based on the same inventive concept, another embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0220] Memory, used to store computer programs;

[0221] A processor, when executing a program stored in memory, implements the resource-optimized virtual power plant control method of the present invention.

[0222] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory can include Random Access Memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0223] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0224] Furthermore, to achieve the above objectives, embodiments of the present invention also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the resource optimization virtual power plant control method of the embodiments of the present invention.

[0225] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable vehicles (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0226] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (apparatus), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0227] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxesFigure 1 The function specified in one or more boxes.

[0228] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0229] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. "" and / or "" indicate that either one or both can be selected. Furthermore, the terms "includes," "contains," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the statement "includes a..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0230] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A resource-optimized virtual power plant control method, characterized in that, The method includes: Collect multi-dimensional monitoring data of the virtual power plant and input it into the virtual power plant state prediction model to perform state prediction and obtain the virtual power plant state prediction results. The virtual power plant state prediction model is constructed based on the STGNN-LSTM-FCI-MLP-PC algorithm, and the virtual power plant state prediction model includes a spatiotemporal feature extraction layer constructed based on the STGNN algorithm, a time series modeling layer constructed based on the LSTM algorithm, a causal inference layer constructed based on the FCI algorithm, a virtual power plant state prediction layer constructed based on the MLP algorithm, and a physical constraint layer constructed based on the PC algorithm, which are connected in sequence. The physical constraint layer includes physical constraint formulas, capacity limitation constraint formulas, and physical feasibility formulas. The physical constraint formula is: In the formula, This is a physical constraint check value; For grid interaction power planning; This represents the actual power exchange between power grids. The virtual power plant state prediction layer outputs the total grid load forecast, photovoltaic power generation forecast, wind power generation forecast, and potential response capability forecast of user-side adjustable load. For power tolerance; t For time indication; Detailed electricity load forecasts for industry, commerce, and residential use; The capacity constraint formula is: In the formula, This is the value for checking capacity limit constraints; For grid interaction power planning; Rated capacity for photovoltaic, wind power, and user-side adjustable loads; These are the minimum and maximum values ​​of the confidence interval for the predicted total load of the power grid; The formula for physical feasibility is: In the formula, This is the value for physical feasibility checks; The total loss function of the virtual power plant state prediction model includes the prediction error loss function and the physical constraint loss function; The formula is: In the formula, This is the total loss function; The prediction error loss function; The physical constraint loss function; Hyperparameters used to balance the weights of physical constraint terms; For other regularization terms; include: Collect multi-dimensional monitoring data of a virtual power plant, preprocess the multi-dimensional monitoring data to obtain preprocessed multi-dimensional monitoring data, and construct the corresponding spatiotemporal map; The spatiotemporal feature extraction layer of the virtual power plant state prediction model is used to extract the spatiotemporal features of the spatiotemporal map. A time series modeling layer using a virtual power plant state prediction model is used to generate time series with spatiotemporal characteristics. Using the causal inference layer of the virtual power plant state prediction model, causal inference results for time series are generated; Based on the time series and causal inference results, the virtual power plant state prediction layer of the virtual power plant state prediction model is used to perform state prediction and obtain the initial virtual power plant state prediction results. The virtual power plant state prediction results include the total grid load prediction value and its confidence interval, the detailed electricity load prediction values ​​of various types of users, the power generation prediction values ​​of various renewable energy sources within the virtual power plant, and the potential response capability prediction values ​​of the user-side adjustable load. The physical constraint layer of the virtual power plant state prediction model is used to physically constrain and correct the initial virtual power plant state prediction results, so as to obtain the final virtual power plant state prediction results. If the physical feasibility check value If the requirements are not met, physical constraints and corrections are necessary. The formula is: In the formula, Correction values ​​for photovoltaic, wind power, and user-side adjustable loads; The function to update the virtual power plant state prediction vector; This is the corrected virtual power plant state prediction vector, which is the final virtual power plant state prediction result. The virtual power plant state prediction results are input into the virtual power plant resource allocation optimization model to generate a resource allocation scheme and obtain the virtual power plant resource allocation scheme. The virtual power plant resource allocation optimization model is constructed based on the HRL algorithm, and the virtual power plant resource allocation optimization model includes a strategic layer and a tactical layer connected in sequence. The strategic layer is equipped with a strategic intelligent agent constructed based on the MILP algorithm, and the tactical layer is equipped with a tactical intelligent agent constructed based on the MPC algorithm. include: Set the global objective and constraints of the virtual power plant resource allocation optimization model, and input the virtual power plant state prediction results into the virtual power plant resource allocation optimization model; The formula is: In the formula, To minimize the optimization objective; The objective function is the global objective function. The total operating cost of the virtual power plant; This represents the total carbon emissions of the virtual power plant. The deviation between the actual output of the virtual power plant and the grid command or predicted net load; This is a penalty for violating the constraints. The global target weight; Based on the virtual power plant status prediction results, the strategic agent of the strategic layer of the virtual power plant resource allocation optimization model is used to optimize resources and obtain the virtual power plant resource operation plan. The formula is: In the formula, Optimize the objective function for the strategic level; To predict the total operating cost of a long-term virtual power plant; To predict the total carbon emissions of a long-term virtual power plant; Optimize target weights for the strategic level; Based on the virtual power plant resource operation plan, the tactical agent of the tactical layer of the virtual power plant resource allocation optimization model is used to generate a resource allocation scheme, thus obtaining the virtual power plant resource allocation scheme. The formula is: In the formula, Optimize the objective function for the tactical layer; The total operating cost of the virtual power plant in the operation plan; The total carbon emissions of the virtual power plant in the operation plan; The deviation between the actual output of the virtual power plant in the operation plan and the grid command or predicted net load; These are penalties for violating constraints in the operational plan, such as equipment exceeding limits or excessive response deviations. Optimize target weights at the tactical level; The virtual power plant resource allocation scheme is input into the virtual power plant digital twin model, and digital twin simulation verification is performed to obtain the corrected virtual power plant resource allocation scheme. The aforementioned virtual power plant digital twin model includes a physical layer, a data layer, a behavioral layer, and a control layer; The revised virtual power plant resource allocation scheme is input into the virtual power plant collaborative control model to generate control commands, obtain virtual power plant control commands, and send them to the corresponding execution terminals. The virtual power plant collaborative control model is based on the MARL algorithm and includes a centralized layer and a decentralized layer. The centralized layer is equipped with a master agent built based on the PPO algorithm, and the decentralized layer is equipped with several sub-agents built based on the DQN algorithm.

2. The resource-optimized virtual power plant control method according to claim 1, characterized in that, The multi-dimensional monitoring data includes power grid-side monitoring data, user-side monitoring data, environmental monitoring data, and resource status monitoring data.

3. The resource-optimized virtual power plant control method according to claim 2, characterized in that, The virtual power plant resource allocation scheme is input into the virtual power plant digital twin model for digital twin simulation verification, resulting in a revised virtual power plant resource allocation scheme, including: Based on the virtual power plant resource allocation scheme and multi-dimensional monitoring data, a digital twin model of the virtual power plant is constructed using digital twin technology, and corresponding simulation scenarios are set up. The virtual power plant resource allocation scheme is input into the virtual power plant digital twin model for dynamic simulation, and key performance indicators are collected in real time during the dynamic simulation process. Based on key performance indicators, digital twin simulation verification is performed to generate corresponding simulation evaluation results; Based on the simulation evaluation results, the virtual power plant resource allocation scheme is dynamically adjusted and optimized to obtain the revised virtual power plant resource allocation scheme.

4. The resource-optimized virtual power plant control method according to claim 3, characterized in that, Based on the virtual power plant resource allocation scheme and multi-dimensional monitoring data, a digital twin model of the virtual power plant is constructed using digital twin technology, and corresponding simulation scenarios are set, including: Based on grid-side monitoring data from multi-dimensional monitoring data, a physical layer is constructed, including several physical models of virtual power plants. Based on real-time data collected from multi-dimensional monitoring data and several historical data collected, a data layer including a data warehouse is constructed. Based on grid-side monitoring data and resource status monitoring data from multi-dimensional monitoring data, a behavior layer is constructed that includes behavioral models of various resources, including virtual power plants. The virtual power plant resource allocation scheme is mapped to the digital twin space to construct a control layer that includes several controllers; The physical layer, data layer, behavior layer, and control layer are integrated into a unified platform to form a virtual power plant digital twin model that is consistent with the topology and resources of the actual virtual power plant. Based on the virtual power plant status prediction results and the virtual power plant resource allocation scheme, a simulation scenario for the virtual power plant digital twin model is set.

5. The resource-optimized virtual power plant control method according to claim 4, characterized in that, The revised virtual power plant resource allocation scheme is input into the virtual power plant collaborative control model to generate control commands, which are then sent to the corresponding execution terminals, including: Based on the resource status monitoring data in the multi-dimensional monitoring data, the various resources of the virtual power plant are allocated to the sub-intelligent agents corresponding to the decentralized layer in the virtual power plant collaborative control model, and the control center of the virtual power plant is allocated to the main intelligent agent corresponding to the centralized layer in the virtual power plant collaborative control model. Set up the corresponding state space, action space, objective function, and global and local information perception capabilities for the main intelligent agent and several sub-intelligent agents, and establish a communication protocol between the main intelligent agent and several sub-intelligent agents. The revised virtual power plant resource allocation scheme is input into the virtual power plant collaborative control model and decomposed into global actions in the action space of the master agent and specific actions in the action space of the sub-agents. Based on the collaborative control mechanism, a distributed optimization algorithm is used to coordinate the control of the main intelligent agent and several sub-intelligent agents, thereby obtaining the execution actions of the sub-intelligent agents. Based on the actions performed by several sub-intelligent agents, control instructions are generated to obtain virtual power plant control instructions, which are then sent to the execution terminal corresponding to each sub-intelligent agent.

6. A resource-optimized virtual power plant control device, used to implement the virtual power plant control method as described in any one of claims 1-5, characterized in that, The device includes: The state prediction unit is used to collect multi-dimensional monitoring data of the virtual power plant, input it into the virtual power plant state prediction model, perform state prediction, and obtain the virtual power plant state prediction result. The resource allocation unit is used to input the virtual power plant state prediction results into the virtual power plant resource allocation optimization model, generate a resource allocation scheme, and obtain the virtual power plant resource allocation scheme. The scheme correction unit is used to input the virtual power plant resource allocation scheme into the virtual power plant digital twin model, perform digital twin simulation verification, and obtain the corrected virtual power plant resource allocation scheme. The collaborative control unit is used to input the modified virtual power plant resource allocation scheme into the virtual power plant collaborative control model, generate control commands, obtain virtual power plant control commands, and send them to the corresponding execution terminals.

Images