Layered and partitioned modeling method and system for virtual power plant, electronic equipment and medium

Abstract

The invention relates to the technical field of electrical engineering and automation, in particular to a layered and partitioned modeling method and system for a virtual power plant, electronic equipment and a medium. The method comprises the following steps: firstly, acquiring a real-time state and multi-subject decision data of distributed energy, and constructing a behavior influence matrix representing a multi-subject interaction relationship based on the real-time state and the multi-subject decision data; constructing a mixed action space model fusing discrete actions and continuous actions based on the matrix, and generating a unified scheduling strategy framework; optimizing the mixed action by adopting an evolutionary algorithm to obtain a multi-region collaborative path; constructing a service interaction network by fusing resource aggregation characteristics and power distribution attributes, and generating a coordinated optimization sequence; and finally, outputting a layered and partitioned coordination model. According to the method, the technical problems of difficulty in coordination of multi-subject benefit games, separation of discrete and continuous action spaces, high cross-regional optimization complexity, poor system adaptability and the like are solved, and the operation reliability, economy and cross-scene migration capability of the virtual power plant are improved.

Description

Technical Field

[0001] This invention relates to the fields of electrical engineering and automation technology, and in particular to a hierarchical and partitioned modeling method, system, electronic equipment, and medium for virtual power plants. Background Technology

[0002] With the deepening of the construction of the energy internet, virtual power plants, by aggregating resources such as distributed wind power, photovoltaics, energy storage, and flexible loads, have become a key means to improve the operating efficiency of the power system and the capacity for renewable energy absorption. They play a crucial role in ensuring power supply stability and promoting the green energy transition. With the deepening of power market reforms and the integration of a high proportion of renewable energy, virtual power plants need to achieve efficient collaborative dispatching across multiple regions and stakeholders under multiple market environments and complex physical constraints to cope with the strong uncertainties and real-time fluctuations on both the source and load sides.

[0003] However, existing virtual power plant optimization scheduling methods still have significant limitations when dealing with scenarios involving multi-market coupling and multi-constraint coordination. Mainstream methods often employ a centralized optimization architecture, assuming all participating entities have aligned goals and complete information transparency, neglecting the inherent heterogeneity of interests and independent decision-making among power generation companies, users, and operators. This simplistic assumption makes it difficult for scheduling models to accurately reflect actual game dynamics. When dealing with large-scale heterogeneous resource aggregation, the generated strategies lack flexibility and cannot adapt to dynamic market signals and real-time physical state changes. Furthermore, existing methods lack effective modeling of cross-regional and cross-level dynamic interactions, making it difficult to balance local autonomy and interest equilibrium in global optimization.

[0004] The core challenge lies in the collaborative modeling of dynamic interactions among multiple stakeholders and the hybrid decision-making space. Virtual power plants involve numerous decision-making entities whose behaviors are interconnected and evolve over time: wind power output adjustments due to electricity price fluctuations in one region may trigger a chain reaction of changes in energy storage dispatch strategies in other regions. Such dynamic correlations are difficult to accurately capture using traditional static optimization methods, and it is even more impossible to simultaneously handle discrete actions (such as equipment start-up and shutdown) and continuous adjustments (such as power allocation) within a unified framework. While reference document 1 (CN112381146B) proposes a distributed resource coordination method based on self-organizing aggregation, achieving resource combination optimization through adaptive stakeholder rules, it relies on predefined aggregation rules and static fitness functions, failing to fully consider the dynamic coupling of real-time market conditions and physical constraints. For example, it fails to address the capacity reuse and physical conservation verification issues of energy storage devices in multi-market transactions. Therefore, when facing minute-level rolling decision-making scenarios, this method still struggles to avoid insufficient strategy feasibility and suboptimal returns.

[0005] Practice has shown that the lack of interactive mechanisms can easily lead to scheduling and coordination failures. A typical scenario is that during peak electricity consumption periods, energy storage in a certain area may fail to respond promptly to a sudden surge in output from neighboring photovoltaic systems, causing localized power shortages or curtailment. Furthermore, in cross-regional collaboration, due to differences in resource characteristics, response speeds, and objectives, unified scheduling commands struggle to achieve efficient cooperation. Especially in scenarios with high penetration rates of renewable energy, regional decision-making biases can trigger a chain of risks such as voltage fluctuations and power exceeding limits, threatening the stable operation of the system.

[0006] Another limitation of existing technologies lies in the lack of a unified decision-making framework for hybrid action spaces. The coexistence of discrete and continuous control variables is a typical characteristic of virtual power plant scheduling, but traditional methods often treat them separately or use heuristic rules for ex-post constraints, resulting in low efficiency and poor feasibility in strategy generation. For example, general optimization algorithms such as mixed-integer linear programming have variable dimensions that grow exponentially with resource scale and number of time periods, failing to meet near real-time clearing requirements; while approximation algorithms based on penalty terms reduce computational complexity, but the sampling results often violate physical constraints, causing oscillations in the training process and easily triggering penalties or execution deviations in practical applications.

[0007] Therefore, the hierarchical and zoned collaborative optimization of virtual power plants still faces the following multiple technical challenges: (1) The real-time perception of multi-source heterogeneous data and the insufficient accuracy of market fluctuation feature extraction affect the reliability of constructing the behavior influence matrix; (2) Deep reinforcement learning has low policy exploration efficiency and poor model robustness in mixed action spaces; furthermore, multi-region collaborative path optimization is prone to getting trapped in local optima, and traditional search methods such as genetic algorithms are difficult to handle high-dimensional dynamic constraints. (3) There is a time mismatch between the coordination mechanism and the market response, and the parameter adjustment is not timely enough; (4) Under the hierarchical structure, it is difficult to synchronize the power balance and benefit distribution across regions, which restricts the overall optimization performance.

[0008] Therefore, the aforementioned problems (1) to (4) highlight the low reliability of existing virtual power plants in actual operation and the difficulty in adapting them to different scenarios. Summary of the Invention

[0009] To address the aforementioned shortcomings or drawbacks, this invention provides a hierarchical and partitioned modeling method, system, electronic equipment, and medium for virtual power plants, which solves the technical problems of low reliability and difficulty in adapting existing virtual power plants to different scenarios during actual operation.

[0010] This invention provides a hierarchical and partitioned modeling method for virtual power plants, comprising: Acquire real-time status and multi-agent decision-making data of distributed energy resources in a virtual power plant, and construct a behavior influence matrix based on the real-time status and multi-agent decision-making data.

[0011] A hybrid action space model is constructed based on the behavior influence matrix, and a unified scheduling strategy framework is generated through the hybrid action space model. The unified scheduling strategy framework includes discrete actions and continuous actions.

[0012] Based on a unified scheduling strategy framework, an evolutionary algorithm is used to optimize discrete and continuous actions to obtain multi-regional collaborative paths.

[0013] Based on multi-regional collaborative paths, a business interaction network is constructed by integrating resource aggregation and power allocation attributes, and a coordinated optimization sequence is generated to drive the operation of the business interaction network.

[0014] The coordinated optimization sequence is mapped to the business interaction network, and a hierarchical partitioned coordination model is output.

[0015] According to a second aspect, the present invention provides a hierarchical and partitioned modeling system for a virtual power plant, comprising: The Behavior Influence Matrix Construction Module is used to acquire real-time status and multi-agent decision-making data of distributed energy in the virtual power plant, and construct a behavior influence matrix based on the real-time status and multi-agent decision-making data.

[0016] The scheduling strategy framework construction module is used to build a hybrid action space model based on the behavior influence matrix, and generate a unified scheduling strategy framework through the hybrid action space model. The unified scheduling strategy framework includes discrete actions and continuous actions.

[0017] The regional collaborative path generation module is used to optimize discrete and continuous actions based on a unified scheduling strategy framework and an evolutionary algorithm to obtain multi-region collaborative paths.

[0018] The collaborative optimization sequence generation module is used to construct a business interaction network based on multi-region collaborative paths, integrating resource aggregation and power allocation attributes, and to generate a collaborative optimization sequence to drive the operation of the business interaction network.

[0019] The hierarchical partitioning model building module is used to map the coordination optimization sequence to the business interaction network and output the hierarchical partitioning coordination model.

[0020] According to a third aspect, the present invention provides an electronic device comprising: At least one processor; and The memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to execute the hierarchical partitioning modeling method of any virtual power plant in the embodiments of the present invention.

[0021] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute a hierarchical partitioning modeling method for any virtual power plant in the embodiments of the present invention.

[0022] The technical solution of this invention first acquires real-time status and multi-agent decision-making data of distributed energy resources in a virtual power plant, and constructs a behavior influence matrix based on the real-time status and multi-agent decision-making data. Next, a hybrid action space model is constructed based on the behavior influence matrix, and a unified scheduling strategy framework is generated through the hybrid action space model. The unified scheduling strategy framework includes discrete actions and continuous actions. Then, based on the unified scheduling strategy framework, an evolutionary algorithm is used to optimize the discrete and continuous actions to obtain multi-regional collaborative paths. Further, based on the multi-regional collaborative paths, a business interaction network is constructed by integrating resource aggregation and power allocation attributes, generating a coordination optimization sequence. This coordination optimization sequence is used to drive the operation of the business interaction network. Finally, the coordination optimization sequence is mapped to the business interaction network to output a hierarchical and partitioned coordination model.

[0023] Throughout the process, this invention addresses the multi-stakeholder conflict of interest problem highlighted in the background technology by dynamically quantifying the decision-making correlation and degree of conflict of interest among power generation companies, users, and grid operators through a behavioral influence matrix, providing a quantitative basis for subsequent coordination mechanisms. Addressing the challenge of coordination in mixed action spaces, it integrates coordinated decision-making for equipment start-up and shutdown and power regulation through a unified scheduling strategy framework, avoiding control failures caused by the fragmented handling of these two types of actions in traditional methods. To address the complexity of cross-regional optimization, it generates a global coordinated path using an evolutionary algorithm while satisfying power balance and transmission constraints, overcoming inter-regional power interaction conflicts caused by local optimization. Finally, to meet the dynamic response requirements of the system, it achieves real-time coupling of resource status and market signals through a business interaction network, ensuring that the coordinated optimization sequence can adapt to environmental changes.

[0024] Therefore, the technical solution of the present invention solves the technical problems of low reliability and difficulty in adapting existing virtual power plants to different scenarios in actual operation, and improves the reliability, economy and cross-scenario adaptability of system-level virtual power plant operation. Attached Figure Description

[0025] Figure 1 This is a flowchart of a hierarchical and partitioned modeling method for a virtual power plant according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a hierarchical and partitioned modeling system for a virtual power plant according to an embodiment of the present invention; Figure 3 This is a block diagram of an electronic device used to implement embodiments of the present invention. Detailed Implementation

[0026] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] This invention provides a hierarchical and partitioned modeling method for virtual power plants, which can be applied to a hierarchical and partitioned modeling system for virtual power plants (hereinafter referred to as the "system"). The physical devices of distributed energy resources in this virtual power plant should be able to run the system's core algorithm module through local deployment or containerization to complete real-time calculation of the hierarchical and partitioned collaborative optimization model and the generation of dynamic scheduling instructions. Specifically, the physical devices of distributed energy resources include, but are not limited to, photovoltaic inverters, wind power generation control systems, energy storage converters, adjustable load controllers, and distributed energy management systems. These devices need to have data acquisition interfaces, communication transmission modules, and strategy execution units to support closed-loop automatic control from data perception and collaborative optimization to instruction issuance. The "hierarchical and partitioned" approach in this invention refers to dividing the optimization control structure of the virtual power plant into multiple levels (such as a centralized scheduling layer, a regional coordination layer, and a local control layer) and multiple regions (such as geographical partitions or electrical characteristic partitions). Through instruction transmission between levels and collaborative interaction between regions, multi-scale management of distributed resources is achieved.

[0028] like Figure 1 As shown, the method may include: Step S110: Obtain the real-time status and multi-agent decision-making data of distributed energy in the virtual power plant, and construct a behavior influence matrix based on the real-time status and multi-agent decision-making data.

[0029] In this context, "real-time status" in real-time status and multi-agent decision-making data refers to physical measurements directly collected by sensors, such as photovoltaic output in kilowatts, energy storage state of charge percentage, and load power in kilowatts. "Multi-agent decision-making data" refers to strategy parameters formulated by different operating entities, such as power generation companies, aggregators, and grid companies, including electricity price adjustment ratios, demand response capacity in megawatt-hours, and safety constraint thresholds. The behavioral impact matrix is ​​a mathematical matrix structure where rows and columns correspond to different decision-making entities and their decision variables. The matrix element values ​​represent the dimensionless impact coefficients (indicating the influence of a change in one entity's decision on the revenue or cost of another entity.

[0030] Specifically, when constructing the behavior influence matrix, the system extracts fluctuation characteristics from market electricity price signals, such as calculating the standard deviation of electricity prices at five-minute intervals; then, the system calculates load adjustment sensitivity from the demand response participation rate; and takes the power generation company's pricing strategy, user-side response parameters, and grid security constraints as inputs to calculate the correlation coefficients between decision variables and fill the matrix.

[0031] For example, if a wind power company increases its output by 10 MW, it may lead to a 0.5% decrease in the revenue of a photovoltaic company in another region (impact coefficient -0.05), or an increase of 0.2 million yuan in grid congestion costs (impact coefficient 0.02). The determination of the impact coefficient is based on the acquired real-time status and multi-entity decision-making data, and is obtained through the following feasible calculation steps: First, based on the market electricity price information, demand response participation rate, and other multi-entity decision-making data, revenue or cost calculation models for different entities are established respectively; then, a baseline scenario is set, and a unit change in a certain entity's decision variable (such as wind power output) is simulated. The calculation model is used to quantify the absolute impact of this change on the revenue or cost of itself and other related entities (such as photovoltaic companies and the power grid). The baseline scenario is defined as the steady-state state of a virtual power plant on a typical operating day, and the parameters include: average load demand (MW), standard electricity price curve (yuan / MWh), and rated operating status of the equipment. Baseline scenario data can be obtained through external electricity market interfaces and energy management systems, and stored in memory for later retrieval. Finally, the absolute impact values ​​are divided by the corresponding entity's revenue or cost baseline values ​​and normalized to obtain dimensionless impact coefficients, which are used to fill the behavioral impact matrix. For example, a coefficient of "-0.05" represents the absolute value of the photovoltaic company's revenue reduction divided by its baseline revenue.

[0032] The revenue or cost calculation model is constructed using linear regression or machine learning algorithms based on market electricity price information and demand response parameters from multi-agent decision-making data. Specifically: Power generation company revenue model: Revenue Where P is the real-time electricity price (yuan / megawatt-hour). Power generation (megawatts). The operating cost (in yuan) includes fuel costs, maintenance expenses, etc., and can be obtained by fitting historical data.

[0033] User cost model: Cost ,in Electricity consumption (megawatts). The penalty cost for demand response is (in yuan), calculated based on interruptible load capacity and response time.

[0034] Grid operator cost model: Cost ,in This is the network loss cost (obtained through power flow calculation). A penalty (in yuan) will be imposed for exceeding safety limits. Model parameters can be calibrated using historical data collected by sensor networks deployed on the distributed energy side, for example, by optimizing the fit using the least squares method.

[0035] Furthermore, to quantify the absolute impact value, the system can simulate a unit change in a certain subject's decision variable (such as an increase of 10 megawatts in wind power output) using a discrete event simulation method: Based on the baseline scenario, modify the values ​​of the target decision variables.

[0036] Run a power flow calculation program (such as one based on the Newton-Raphson method) to update the system power distribution.

[0037] The updated status is input into the revenue / cost model to calculate the change in revenue or cost for each entity, i.e., the absolute impact value (unit: yuan). For example, an increase of 10 megawatts in wind power output may lead to a decrease in revenue for photovoltaic companies. Yuan, power grid costs increased Yuan. The absolute impact value is calculated in real time using simulation tools (such as DigSILENT or a custom algorithm).

[0038] Step S120: Construct a hybrid action space model based on the behavior influence matrix, and generate a unified scheduling strategy framework through the hybrid action space model.

[0039] The unified scheduling strategy framework includes discrete actions and continuous actions.

[0040] Specifically, the hybrid action space model is a mathematical model used to uniformly describe discrete control commands (such as equipment start-up and shutdown, and switch switching) and continuous regulation commands (such as power allocation ratios and output setpoints). The unified scheduling strategy framework is a set of decisions containing all possible action combinations, and its output is an executable sequence of scheduling commands.

[0041] For example, the system defines the discrete action space as a binary variable, where 0 represents shutdown and 1 represents startup; then, the continuous action space is defined as a real variable (such as a power allocation ratio of 0.1 to 1.0); finally, the system uses a neural network model (such as MLP, which stands for Multilayer Perceptron) to map the two types of actions to a unified representation space, generating a scheduling instruction table containing timestamps, region identifiers, and action types.

[0042] Step S130: Based on the unified scheduling strategy framework, an evolutionary algorithm is used to optimize discrete and continuous actions to obtain multi-regional collaborative paths.

[0043] Among them, the multi-regional collaborative path is a time-series action sequence that specifies the power exchange plan, equipment switching status and adjustment amount of resources in each region within different time periods.

[0044] In this embodiment, the optimization objective can be to minimize the total operating cost (ten thousand yuan) and maximize the renewable energy absorption rate (percentage). The constraints can include the line transmission capacity (megawatts), voltage deviation percentage, and equipment ramp rate (megawatts per minute).

[0045] Specifically, the system initializes the population (size 100) using a genetic algorithm (GA). Each chromosome encodes discrete action gene segments (binary strings) and continuous action gene segments (real number arrays). The fitness function comprehensively calculates economic indicators (such as a 10% cost reduction) and safety indicators (such as a 5-fold reduction in voltage exceedances). When performing crossover operations, the system can use single-point crossover (for the discrete part) and arithmetic crossover (for the continuous part). When performing mutation operations, the system can use bit flipping (for the discrete part) and Gaussian perturbation (for the continuous part, with a standard deviation of 0.1).

[0046] Step S140: Based on the multi-regional collaborative path, integrate resource aggregation and power allocation attributes to construct a service interaction network, and generate a coordinated optimization sequence to drive the operation of the service interaction network.

[0047] Among them, building a business interaction network refers to the system establishing a multi-layer network architecture based on resource aggregation attributes and power allocation attributes, which integrates physical layer device connection, information layer data interaction and business layer protocol collaboration, so as to realize dynamic coordination and optimization decision-making across regions and entities within the virtual power plant; the coordination optimization sequence is a set of instructions arranged in chronological order. The coordination optimization sequence specifies the coordinated operation scheme of discrete actions (such as equipment start-up and shutdown) and continuous actions (such as power allocation) that each region's resources need to perform within a specific time period.

[0048] Specifically, the system can complete the construction of the business interaction network and the generation of the coordination and optimization sequence through the following steps: Step 1: Based on the regional boundaries and resource topology relationships defined by the multi-regional collaborative path, establish a device connection mapping table at the physical layer (recording the correspondence between nodes and distributed energy resources), configure data communication protocols at the information layer (such as binding logical devices and data attributes using the IEC61850 protocol), and set collaborative rules at the business layer (such as a cross-regional power deviation responsibility sharing coefficient of 0.6). Step 2: Input the resource aggregation characteristics (such as the equivalent response rate of the energy storage cluster of 2 MW / min and the adjustable capacity range of 0 to 50 MW) and power allocation attributes (such as the line transmission loss coefficient of 0.05 and the node voltage safety limit of 0.95 to 1.05 per unit) into the linear programming model, solve for the optimal power allocation scheme that satisfies the cross-regional power balance constraints (such as the net exchange power deviation ≤1%), and verify the physical feasibility of the scheme through power flow calculation. Step 3: Based on the verified power allocation scheme, generate a coordinated optimization sequence according to a preset time granularity (e.g., 15 minutes). For example: During time period T1, the energy storage charging power of region A is set to 10 MW (continuous action), and the photovoltaic inverter start / stop status of region B is set to 1 (discrete action); During time period T2, the power transmitted from region A to region B is adjusted to 20 MW (continuous action), and the load shedding command of region C is triggered (discrete action); This sequence is sent to each region controller through the information layer communication interface of the service interaction network, driving the physical layer equipment to perform the corresponding operations.

[0049] Step 4: Collect the device execution results (such as the error between actual power and command) in real time through the status monitoring module in the service interaction network. If the deviation exceeds the set threshold (such as 5%), the protocol adaptation module is triggered to adjust the subsequent sequence parameters (such as correcting the transmission power value or rearranging the device action timing) to form a closed-loop optimization mechanism.

[0050] For example, resource aggregation attributes may include the equivalent adjustment rate of the distributed resource cluster in megawatts per minute, the maximum adjustable capacity in megawatts, and the minimum continuous operating time in minutes; power allocation attributes may include the percentage of line transmission loss, the node voltage limit in kilovolts, and the power flow safety margin in megawatts.

[0051] For example, the system uses a clustering algorithm (K-means) to group resources according to their characteristics, forming multiple virtual adjustable units; and uses linear programming to calculate the optimal power allocation ratio to ensure that the power balance deviation between regions is less than 1 megawatt.

[0052] For example, the coordinated optimization sequence is formatted as a JSON structure, containing a time window (e.g., 2023-08-01T10:00:00 to 2023-08-01T10:15:00), a 50 MW transmission plan from region A to region B, and a 30 MW energy storage charging instruction. The sequence generation cycle is consistent with the market clearing cycle (15 minutes), and digital signatures ensure the instructions are tamper-proof.

[0053] Step S150: Map the coordination optimization sequence to the business interaction network and output a hierarchical partition coordination model.

[0054] Among them, the hierarchical partition coordination model is a digital model that includes physical layer topology, information layer communication protocol, and business layer scheduling rules, used to describe the multi-level collaborative operation mechanism of virtual power plants.

[0055] For example, the physical layer model includes region boundary coordinates and line impedance parameters in ohms; the information layer model includes data sampling frequency in Hertz and communication delay in milliseconds; and the service layer model includes cost allocation rules and deviation assessment coefficients in yuan per megawatt-hour. The model output format is a standardized configuration file (e.g., YAML format, a user-friendly data serialization standard format), which supports importing into third-party simulation platforms (such as MATLAB / Simulink) for verification.

[0056] For example, model version V2.1.3 supports a dynamic update mechanism, automatically expanding partition boundaries when new distributed resources are detected. This dynamic update mechanism is implemented through the following techniques: The system periodically (e.g., every 24 hours) calls the resource registration interface to scan for new distributed resources. When a resource change is detected, a version verification process is triggered. First, the compatibility of the new resource is verified based on resource aggregation characteristics (e.g., response rate, adjustable capacity range). If the verification passes, a new version number (e.g., V2.1.4) is automatically generated. The version number is obtained by pulling an incrementing identifier from the central version management server or by dynamically calculating it based on a timestamp and resource hash value. Subsequently, the configuration file (e.g., YAML format) of the hierarchical partition coordination model is updated, and the changes are synchronized in real time through the status monitoring module in the business interaction network to ensure that partition boundary adjustments comply with power balance constraints and safety limits. This mechanism enables the model to adapt to resource changes, improving system implementability.

[0057] Therefore, according to the above implementation method, firstly, real-time status and multi-agent decision-making data of distributed energy resources in the virtual power plant are acquired, and a behavior influence matrix is ​​constructed based on the real-time status and multi-agent decision-making data. Next, a hybrid action space model is constructed based on the behavior influence matrix, and a unified scheduling strategy framework is generated through the hybrid action space model. The unified scheduling strategy framework includes discrete actions and continuous actions. Then, based on the unified scheduling strategy framework, an evolutionary algorithm is used to optimize discrete and continuous actions to obtain multi-regional collaborative paths. Further, based on the multi-regional collaborative paths, a business interaction network is constructed by integrating resource aggregation and power allocation attributes, generating a coordination optimization sequence. This coordination optimization sequence is used to drive the operation of the business interaction network. Finally, the coordination optimization sequence is mapped to the business interaction network, outputting a hierarchical and partitioned coordination model.

[0058] Throughout the process, this embodiment addresses the multi-stakeholder conflict of interest problem highlighted in the background technology by dynamically quantifying the decision-making correlation and degree of conflict of interest among power generation companies, users, and grid operators through a behavioral influence matrix, providing a quantitative basis for subsequent coordination mechanisms. To address the challenge of coordination in mixed action spaces, a unified scheduling strategy framework integrates coordinated decisions for equipment start-up and shutdown (discrete actions) and power regulation (continuous actions), avoiding control failures caused by the fragmented handling of these two types of actions in traditional methods. To address the complexity of cross-regional optimization, an evolutionary algorithm generates a global coordinated path while satisfying power balance and transmission constraints, overcoming inter-regional power interaction conflicts caused by local optimization. Finally, to meet the system's dynamic response requirements, a business interaction network enables real-time coupling of resource status and market signals, ensuring that the coordinated optimization sequence can adapt to environmental changes.

[0059] Therefore, the technical solution of this embodiment solves the technical problems of low reliability and difficulty in adapting existing virtual power plants to different scenarios in actual operation, and improves the reliability, economy and cross-scenario adaptability of system-level virtual power plant operation.

[0060] In some embodiments, real-time status and multi-agent decision-making data include distributed energy operation status data, environmental parameters, market electricity price information, demand response participation rate, electricity price adjustment parameters, demand response parameters, and security constraint parameters; the step of acquiring real-time status and multi-agent decision-making data of distributed energy in a virtual power plant includes: Operational status data is collected through a sensor network deployed on the distributed energy side.

[0061] Specifically, the operating status data may include the real-time power output in kilowatts of the photovoltaic inverter, the state of charge percentage of the energy storage system, and the operating temperature in degrees Celsius of the fuel cell.

[0062] Environmental parameters are collected through environmental monitoring units deployed on the distributed energy side.

[0063] The environmental parameters may specifically include irradiance in watts per square meter, wind speed in meters per second, and ambient temperature in degrees Celsius.

[0064] We obtain market electricity price information and demand response participation rate through external electricity market data interfaces.

[0065] Among them, market electricity price information may include the clearing price per megawatt-hour in both the day-ahead market and the real-time market, and the demand response participation rate refers to the percentage of actual response load to total adjustable load.

[0066] Electricity price adjustment parameters are obtained through the decision-making system of external power generation companies, and safety constraint parameters are obtained through the interface of external power grid dispatch center.

[0067] Specifically, the safety constraint parameters may include the slope of the price curve for gas turbine units and the subsidy coefficient per kilowatt-hour for new energy power plants; and the safety constraint parameters may be obtained through the external power grid dispatch center interface, including the line transmission capacity limit in megawatts, the allowable percentage deviation of node voltage, and the system spinning reserve requirement in megawatts.

[0068] Demand response parameters are obtained through the user-side energy management system.

[0069] Specifically, demand response parameters may include interruptible load capacity in kilowatts, minimum response duration in minutes, and load recovery rate in kilowatts per minute.

[0070] Therefore, according to the above implementation method, the system can achieve standardized access to multi-source heterogeneous data, providing a complete, consistent, and traceable data foundation for constructing a behavior influence matrix. For example, a virtual power plant includes three types of resources: photovoltaic power plants, energy storage systems, and industrial loads. The system acquires photovoltaic power curves, energy storage charging and discharging status, and real-time load usage through sensors; obtains time-of-use electricity price signals through market interfaces; and obtains line safety limits through the power grid dispatch system. Ultimately, a unified dataset containing numerical, state, and constraint data is formed, effectively supporting subsequent collaborative optimization calculations.

[0071] In some embodiments, the step of constructing a behavior influence matrix based on real-time status and multi-agent decision-making data includes: Extract price fluctuation characteristics from market electricity price information.

[0072] Specifically, the price fluctuation characteristics can include calculating the standard deviation (reflecting the intensity of fluctuation) and the root mean square value (reflecting the amplitude of fluctuation) of the hourly electricity price, both in yuan / megawatt-hour.

[0073] Furthermore, load response characteristics are extracted from the demand response participation rate, and price fluctuation characteristics are combined with load response characteristics to form a market fluctuation characteristic set.

[0074] Specifically, the market volatility feature set may include load adjustment rate (unit: megawatts / minute) and response duration (unit: minutes); the system combines price volatility features and load response features to form the market volatility feature set, which is a vector set containing multiple quantitative indicators.

[0075] Next, based on market fluctuation characteristic sets, electricity price adjustment parameters (e.g., the slope of the power generation company's price quotation curve), demand response parameters (e.g., interruptible load capacity on the user side), and safety constraint parameters (e.g., line transmission capacity limits), the system uses a multiple linear regression method to calculate the dimensionless weight of each subject's decision variables relative to the revenue of other subjects. This weight represents the degree of influence of a subject's decision change on the revenue of another subject. The implementation of the multiple linear regression method includes the following steps: First, define the dependent and independent variables of the regression model, where the revenue change of a specific subject is used as the dependent variable, and the decision variables of other relevant subjects (key quantitative indicators extracted from market fluctuation characteristic sets, electricity price adjustment parameters, demand response parameters, and safety constraint parameters, such as price fluctuation standard deviation, interruptible load capacity, and line transmission capacity limits) are used as independent variables; then, collect historical operating data or generate sample datasets through simulation to ensure that the samples cover multiple typical operating scenarios; subsequently, use the least squares method to fit the multiple linear regression equation, in the form of... , where ΔY represents the dependent variable (change in earnings). The independent variable (decision variable indicator). This is the initial influence weight of each decision variable relative to the subject's return; finally, the regression coefficient βᵢ is normalized by dividing it by the absolute value of the dependent variable's baseline value, and the result is constrained within the interval [-1, 1] to finally obtain the dimensionless return influence weight.

[0076] An initial behavioral influence matrix is ​​constructed based on the weights of each benefit. Each element in the initial behavioral influence matrix is ​​used to characterize the correlation coefficient between decision variables.

[0077] Specifically, the system can construct an initial behavioral impact matrix based on the weights of each benefit. This matrix is ​​a square matrix, with rows and columns corresponding to different decision-making entities, such as power generation companies, users, and grid operators. Each element in the initial behavioral impact matrix is ​​used to characterize the correlation coefficient between decision variables (range of values). Positive values ​​indicate positive effects, while negative values ​​indicate negative effects.

[0078] Time series analysis was used to correlate environmental parameters with the real-time total power output of the virtual power plant in order to update the dynamic coupling parameters in the initial behavior influence matrix.

[0079] The real-time total power output of the virtual power plant refers to the algebraic sum of the net power injected into or absorbed from the grid by all distributed energy sources, such as photovoltaics, wind power, energy storage, and adjustable loads, within the jurisdiction of the virtual power plant at a certain moment. The value of this algebraic sum is equal to the instantaneous difference between the total power generation and the total power consumption. The grid refers to the upper-level transmission and distribution network connected to the virtual power plant.

[0080] Specifically, the system can use time series analysis methods, such as the autoregressive integral moving average model ARIMA, to perform correlation analysis between environmental parameters (temperature, irradiance) and the real-time total power output (unit: megawatt) of the virtual power plant in order to calculate the sensitivity coefficient of environmental factors to power output, and update the dynamic coupling parameters in the initial behavior influence matrix accordingly (i.e., adjust the matrix element values ​​to reflect the correction amount of the impact of environmental changes on decision-making).

[0081] The feasibility of the updated initial behavior impact matrix is ​​verified using runtime status data, and the behavior impact matrix is ​​output after the feasibility verification is completed.

[0082] Specifically, the system can use operational status data, such as device availability and power limit alarms, to perform a feasibility check on the updated initial behavior impact matrix. This includes checking whether the matrix elements meet physical constraints (such as power balance), logical constraints (such as weight sign consistency), and stability constraints (such as eigenvalue magnitude less than 1), and outputting the final behavior impact matrix after the feasibility check is completed.

[0083] Therefore, based on the above implementation method, the system can construct a dynamic matrix that accurately quantifies the interactive relationships among multiple entities, providing core decision-making basis for the collaborative optimization of virtual power plants. For example, in a certain scenario, if the grid operator adjusts the transmission price (decision variable) by 0.1 yuan / kWh, the behavioral impact matrix can predict that the power generation company's revenue will decrease by 50,000 yuan (impact weight of -0.5), and the user's electricity purchase cost will increase by 30,000 yuan (impact weight of -0.3), thus providing quantitative support for collaborative dispatch.

[0084] In some embodiments, the step of constructing a hybrid action space model based on a behavior influence matrix includes: Extract the coupling relationships and conflict weights among multi-agent decision variables from the behavior influence matrix to determine the spatial boundary constraints of discrete and continuous actions.

[0085] Among them, spatial boundary constraints refer to the feasible domain range set for discrete and continuous actions in the mixed action space, such as the mutual exclusion constraint of discrete actions (the same device cannot execute start and stop commands at the same time) and the upper and lower limit constraints of continuous actions (the power allocation ratio must be between 0 and 1).

[0086] The equipment start / stop status and switch switching commands are defined as discrete action space, while the power distribution ratio and output adjustment amount are defined as continuous action space.

[0087] The discrete action space consists of a finite number of discrete control commands, such as the start / stop status of the photovoltaic inverter (0 indicates off, 1 indicates on) and the switching of the working mode of the energy storage converter (0 indicates charging, 1 indicates discharging); the continuous action space consists of real variables, such as the wind power allocation ratio (0.1 indicates 10% of the rated power) and the gas turbine output adjustment (±50 MW).

[0088] The spatial boundary constraints are input into a pre-defined neural network model, and joint representation and dimension fusion operations are performed on the discrete action space and the continuous action space to output a hybrid action representation vector.

[0089] Specifically, the neural network model adopts a multilayer perceptron (MLP) structure, including an input layer (receiving boundary constraint parameters), a hidden layer (performing nonlinear transformations), and an output layer (generating a hybrid action representation vector). Dimension fusion refers to merging the one-hot encoding of discrete actions with the normalized values ​​of continuous actions into a unified vector through a fully connected layer; for example, merging discrete action [0,1] and continuous action [0.5] into vector [0,1,0.5]. The neural network model, using a multilayer perceptron (MLP) structure, has the following core parameter configurations: the dimension of the input layer is determined by the total number of spatial boundary constraint parameters and is used to receive the normalized constraint parameters; the hidden layer contains at least one fully connected layer, with the number of neurons configurable from 64 to 256, and the number of layers can be set from 1 to 3 depending on the problem complexity; the activation function uses the ReLU function to introduce a nonlinear transformation; the dimension of the output layer is consistent with the dimension of the hybrid action representation vector, the number of neurons in the activation function is equal to the sum of the dimensions of the discrete action space and the continuous action space, and the activation function uses a linear function to ensure the continuous range of the output values. The MLP model is trained through supervised learning, with the training objective being to minimize the mean square error between the predicted action representation and the optimal action baseline.

[0090] A set of state-action mapping rules is constructed based on the dynamic coupling parameters in the hybrid action representation vector and the behavior influence matrix.

[0091] The state-action mapping rule set is a set of conditional rules that specify the combination of mixed actions to be executed under a specific system state (such as a load demand of 100 MW and an electricity price of 50 yuan / MWh). For example, "If the node voltage is lower than the rated value of 0.95, the energy storage discharge action (discrete action) should be initiated first, and at least 20% of the power should be allocated (continuous action)." Based on the state-action mapping rule set, the initial scheduling policy set is generated using the policy gradient algorithm.

[0092] Specifically, the policy gradient algorithm optimizes policy parameters through gradient ascent, for example, by using the REINFORCE algorithm to update the weights of a neural network; the initial scheduling policy set contains multiple candidate policies, each of which is a sequence of actions in a time series, such as policy A below: [Start / stop status at time t1 = 1, power allocation = 0.3; Start / stop status at time t2 = 0, power allocation = 0.6].

[0093] The initial scheduling strategy set is subjected to action conflict resolution and feasibility verification, and a hybrid action space model is constructed based on the initial scheduling strategy set after action conflict resolution and feasibility verification.

[0094] Among them, action conflict resolution refers to detecting and eliminating action combinations in the strategy that violate physical logic, such as the frequent start-up and shutdown of the same equipment in adjacent time periods (which need to be merged into continuous operation); when the system performs feasibility verification, it can perform operations such as verifying power balance (total power generation and total load error is less than 1%) and equipment capacity limitation (output does not exceed the rated value); the final output hybrid action space model is the set of verified scheduling strategies and their corresponding state action mapping rules.

[0095] Therefore, according to the above implementation method, the system can achieve unified modeling and collaborative optimization of discrete and continuous control commands. For example, a virtual power plant includes photovoltaic, energy storage, and load resources. The model can simultaneously generate joint strategies for "photovoltaic start-up and shutdown (discrete action)" and "energy storage power allocation (continuous action)," avoiding control conflicts caused by the separation of the two types of actions in traditional methods, and improving the feasibility and economy of the scheduling scheme.

[0096] In some embodiments, based on a unified scheduling strategy framework, an evolutionary algorithm is used to optimize discrete and continuous actions to obtain multi-regional cooperative paths, including: Power balance constraints, transmission capacity constraints, and response time constraints are extracted from the unified scheduling strategy framework to form a set of constraints.

[0097] Among them, power balance constraint refers to the physical condition that regional power generation and load demand must be matched in real time. For example, the total power generation of region A (100 MW) must be equal to the total load power (95 MW) plus network loss (5 MW); transmission capacity constraint refers to the upper limit of power that a line can transmit. For example, the maximum transmission capacity of tie line AB is 50 MW; response time constraint refers to the control command that must be executed within a specified time. For example, the energy storage power station must reach the specified output within 5 minutes.

[0098] An initial population is generated based on the set of constraints. The initial population includes multiple candidate cooperative paths. Each candidate cooperative path includes a spatial sequence of discrete actions and parameterized instructions for continuous actions.

[0099] Specifically, the initial population is generated randomly or through heuristic rules, typically ranging from 50 to 200 individuals; candidate collaborative paths are action plans extending from the time to the space dimension, for example, path X includes: [time period] Area 1 Photovoltaic Start / Stop = 1 (Activated); Area 2 Energy Storage Power Allocation = 0.3 (30% of Rated Power); Time Period :Region 1 Photovoltaic Start-up / Stop = 0 (Off), Region 2 Energy Storage Power Allocation = 0.6.

[0100] Based on the benefit impact weight and conflict weight in the behavior impact matrix, and combined with the constraint set, the comprehensive fitness of each candidate collaborative path is calculated.

[0101] Specifically, the system can calculate the overall fitness using the following weighted summation formula: Economic benefits are calculated based on electricity prices and power generation, in tens of thousands of yuan. Constraint violation penalties quantify power deviation values ​​(in megawatts); conflict costs are dimensionless parameters derived from conflict weights in the behavior impact matrix; weighting coefficients can be: =0.6, =0.3, =0.1 (adjustable); the penalty for constraint violation refers to the degree of exceeding the limit (such as power deviation in megawatts), and the conflict cost is quantified according to the conflict weight in the behavior impact matrix (dimensionless).

[0102] The initial population is iteratively evolved using selection, crossover, and mutation operations. Crossover operations include fragment recombination of discrete action sequences and arithmetic crossover of continuous action parameters. Mutation operations include random flipping of discrete actions and Gaussian perturbation of continuous actions.

[0103] Specifically, when performing selection operations, the system can use a roulette wheel selection method, where individuals with higher fitness have a greater probability of being selected; when performing fragment recombination, the system can swap discrete action sequences of the same time period between two individuals, for example, swapping the start and stop instructions of time periods 10 to 20 in paths A and B; arithmetic crossover performs linear interpolation on continuous parameters. ; The random flip reverses the discrete action bits (0 becomes 1 or 1 becomes 0) with a probability p (e.g., 0.05). Next, the system can add noise to the continuous parameters using Gaussian perturbations: ; Wherein, the standard deviation σ = 0.1 × parameter range.

[0104] When the maximum number of iterations is reached, or when the overall fitness of each candidate cooperative path converges to a set threshold, the candidate cooperative path with the highest overall fitness is selected from the current population as the optimized multi-region cooperative path.

[0105] Specifically, when the maximum number of iterations is reached, such as 1000 generations, or when the overall fitness of each candidate cooperative path converges to a set threshold (e.g., fitness improvement is less than 1% for 10 consecutive generations), the system selects the candidate cooperative path with the highest overall fitness from the current population as the optimized multi-region cooperative path.

[0106] Therefore, according to the above implementation method, the system can efficiently search for global optimization solutions that satisfy complex constraints through evolutionary algorithms. For example, a virtual power plant includes 3 regions and 5 types of resources. The algorithm can generate a 24-hour scheduling plan within 10 minutes, ensuring that the total revenue is maximized (expected to increase revenue by 5%) while fully meeting the grid security constraints, thus improving the economy and reliability of cross-regional collaboration.

[0107] In some embodiments, the step of constructing a service interaction network based on multi-regional collaborative paths and integrating resource aggregation and power allocation attributes includes: Extract cross-regional power exchange plans and equipment scheduling instructions from multi-regional collaborative paths to generate an initial resource allocation scheme.

[0108] Among them, the cross-regional power exchange plan refers to the planned power value and direction between different regions, such as region A transmitting 50 MW to region B; equipment scheduling instructions include discrete actions (such as enabling the charging mode of the energy storage power station) and continuous actions (such as setting the output power of the photovoltaic power station to 80 MW).

[0109] Based on the resource aggregation characteristics, the schedulable capacity of the initial resource allocation scheme is verified.

[0110] The resource aggregation characteristics include the response rate of distributed resources (unit: megawatts / minute), adjustable capacity range (unit: megawatts), and minimum dwell time (unit: minutes). Specifically, during the verification process, the system calculates whether the aggregated resources can reach the commanded power within the required time (e.g., a 100-megawatt load must be responded to within 5 minutes) and ensures that the duration of the action is not less than the minimum operating time of the equipment (e.g., the minimum operating time of a gas turbine is 30 minutes).

[0111] Based on the set power allocation attributes, the verified resource allocation scheme is optimized for regional power balance. The power allocation attributes include transmission loss coefficient, node voltage limit and power flow safety constraints.

[0112] Specifically, the system can optimize regional power balance of the verified resource allocation scheme based on power allocation attributes, including transmission loss coefficient (unit: percentage), node voltage limit (unit: kilovolt), and power flow safety constraint (unit: megawatt). The optimization objective is to minimize total network loss and ensure that node voltage deviation does not exceed ±5%.

[0113] The formula for minimizing total network loss is as follows: .

[0114] The optimized power allocation scheme is used to detect conflicts of interest and make balance adjustments based on the benefit impact weights in the behavior impact matrix.

[0115] Specifically, when performing conflict of interest detection, the system identifies conflicting benefits between different stakeholders (such as increased revenue for power generation companies potentially leading to higher costs for users), and uses a weighted compromise algorithm for equilibrium adjustment. ; Where α is the compromise coefficient, which is set to 0.7 by default.

[0116] Create a multi-layer network topology, which includes the physical layer, information layer, and service layer.

[0117] The physical layer describes electrical connections (such as node impedance matrices), the information layer defines communication protocols and data formats (such as the IEC 61850 protocol), and the business layer contains scheduling rules and market mechanisms (such as clearing algorithms).

[0118] The adjusted resource allocation scheme is mapped to the physical layer, the optimized power allocation scheme is mapped to the information layer, and the status monitoring module and protocol adaptation module are integrated into the service layer of the multi-layer network topology to generate a service interaction network.

[0119] Specifically, the system can map the adjusted resource allocation scheme to the physical layer as device control commands, then map the optimized power allocation scheme to the information layer as a communication data source, and integrate a status monitoring module (for real-time acquisition of device status) and a protocol adaptation module (supporting multi-protocol conversion) in the service layer to generate a service interaction network.

[0120] Therefore, according to the above implementation method, the system can achieve deep integration of physical operation and business management. For example, a power transmission scheme between regions is executed through a business interaction network: the physical layer controls the energy storage power station to output 80 MW, the information layer synchronously pushes power data to the dispatch center, and the business layer dynamically adjusts the transmission plan according to market prices, ultimately ensuring that cross-regional collaboration meets both physical security constraints and achieves economic optimization.

[0121] In some embodiments, the coordination optimization sequence is mapped to the service interaction network, outputting a hierarchical partition coordination model, including: Cross-layer interest adjustment parameters and partitioned coordination protocols are extracted from the coordination optimization sequence to generate a hierarchical coordination rule set.

[0122] Among them, the cross-level benefit adjustment parameter refers to the benefit distribution coefficient (dimensionless) between different levels of entities (such as regional dispatch centers and distributed resource clusters), for example, the regional level benefits account for 60% and the cluster level benefits account for 40%; the partitioned coordination protocol refers to the constraint rules for power exchange between regions (such as a maximum exchange power of 50 megawatts and a minimum exchange time of 15 minutes).

[0123] Based on the state monitoring and protocol adaptation results in the business interaction network, the hierarchical coordination rule set is dynamically modified to obtain the updated hierarchical coordination rules.

[0124] The status monitoring results can include real-time equipment operating status (e.g., energy storage state of charge dropping to 30%), network power flow limit alarms (e.g., line load rate exceeding 90%), and protocol adaptation results (e.g., data conversion power (e.g., 98.5%) for different communication protocols (e.g., IEC 61850, Modbus). The correction method uses the following sliding window weighted average: .

[0125] Based on the updated hierarchical coordination rules, the partitioned topology of the virtual power plant is reconstructed, including the partition boundary adjustment mechanism and cross-layer power balance constraints.

[0126] Among them, the partition boundary adjustment mechanism refers to the system re-dividing electrical areas according to the dynamic aggregation results of resources, such as transferring a certain energy storage cluster in the original area A to area B; the cross-layer power balance constraint is used to require that the difference between power generation and load at each level be within the allowable range (such as area-level deviation ±2%, cluster-level deviation ±5%).

[0127] The reconstructed partition topology is mapped to the physical and information layers of the business interaction network to generate a basic framework for partition coordination.

[0128] As described in the above embodiments, physical layer mapping may include updating the node admittance matrix (reflecting the electrical parameters of the new topology), and information layer mapping may include adjusting the configuration of data acquisition points (such as adding 10 measurement points in region B); the generated partition coordination basic framework includes topology connection relationships, communication routing tables, and data flow definitions.

[0129] A hierarchical decision engine is built based on the partition coordination framework and the updated hierarchical coordination rules.

[0130] The hierarchical decision engine can be composed of a regional autonomous scheduling module (handling intra-region optimization) and a cross-regional collaborative decision-making module (handling inter-regional power exchange), and uses a distributed optimization algorithm (such as the Alternating Direction Multiplier Method, ADMM) to achieve parallel solution. The ADMM algorithm decomposes the optimization problem into subproblems to achieve inter-regional collaboration. In specific applications, a penalty parameter needs to be set to balance convergence speed and accuracy.

[0131] The hierarchical decision engine executes a coordination optimization sequence and outputs a hierarchical partition coordination model.

[0132] Specifically, when the system executes the optimization sequence, it first parses the instructions in the optimization sequence, such as "region A transmits 30 megawatts to region B"; then it calls the corresponding module to calculate the optimal scheduling strategy, and finally outputs a network-wide coordination model that includes partition control instructions, cross-region exchange plans and security verification results.

[0133] Therefore, according to the above implementation method, the system can achieve dynamic collaborative optimization across levels and multiple regions. For example, when the photovoltaic output of a certain region drops sharply, the system quickly adjusts the partition topology (incorporating backup energy storage into the region) and updates the power allocation scheme (transferring 20 MW from a neighboring region) through the hierarchical decision engine, restoring power balance within 5 minutes and improving the reliability and response speed of the virtual power plant to emergencies.

[0134] Figure 2 This is a structural block diagram of a hierarchical and partitioned modeling system for a virtual power plant according to an embodiment of the present invention.

[0135] like Figure 2 As shown, the hierarchical and partitioned modeling system for this virtual power plant includes: The behavior influence matrix construction module 210 is used to acquire the real-time status and multi-agent decision data of distributed energy in the virtual power plant, and construct a behavior influence matrix based on the real-time status and multi-agent decision data. The scheduling strategy framework construction module 220 is used to construct a hybrid action space model based on the behavior influence matrix, and generate a unified scheduling strategy framework through the hybrid action space model. The unified scheduling strategy framework includes discrete actions and continuous actions. The regional collaborative path generation module 230 is used to optimize the discrete and continuous actions using an evolutionary algorithm based on the unified scheduling strategy framework to obtain a multi-regional collaborative path. The collaborative optimization sequence generation module 240 is used to construct a service interaction network based on the multi-region collaborative path, integrate resource aggregation and power allocation attributes, and generate a collaborative optimization sequence to drive the operation of the service interaction network. The hierarchical partitioning model construction module 250 is used to map the coordination optimization sequence to the business interaction network and output the hierarchical partitioning coordination model.

[0136] The specific functions and examples of each module and submodule of the device in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0137] According to embodiments of the present invention, the above-described method of the present invention can be applied to an electronic device and a readable storage medium.

[0138] Figure 3 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0139] like Figure 3 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0140] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0141] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a hierarchical partitioning modeling method for a virtual power plant. For example, in some embodiments, a hierarchical partitioning modeling method for a virtual power plant can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the hierarchical partitioning modeling method for a virtual power plant described above can be performed. Alternatively, in other embodiments, computing unit 601 may be configured by any other suitable means (e.g., by means of firmware) to perform a hierarchical partitioning modeling method for a virtual power plant.

[0142] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0143] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0144] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0145] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual, auditory, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0146] The systems and technologies described herein can be implemented in computing systems that include back-end components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0147] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0148] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0149] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A hierarchical zoned modeling method of a virtual power plant, characterized in that, The method comprises the following steps: acquiring real-time state and multi-agent decision data of distributed energy in a virtual power plant, and constructing a behavior influence matrix according to the real-time state and multi-agent decision data; constructing a hybrid action space model based on the behavior influence matrix, and generating a unified scheduling strategy framework through the hybrid action space model, wherein the unified scheduling strategy framework comprises discrete actions and continuous actions; based on the unified scheduling strategy framework, optimizing the discrete actions and continuous actions by using an evolutionary algorithm to obtain a multi-region collaborative path; based on the multi-region collaborative path, constructing a business interaction network by fusing resource aggregation and power distribution attributes, and generating a coordination optimization sequence for driving the business interaction network to run; mapping the coordination optimization sequence to the business interaction network, and outputting a hierarchical partition coordination model.

2. The method of claim 1, wherein, The real-time state and multi-agent decision data comprise operating state data of the distributed energy, environmental parameters, market electricity price information, demand response participation rate, electricity price adjustment parameters, demand response parameters and safety constraint parameters; The step of acquiring the real-time state and multi-agent decision data of the distributed energy in the virtual power plant comprises: collecting the operating state data through a sensor network deployed on the side of the distributed energy, and collecting the environmental parameters through an environmental monitoring unit deployed on the side of the distributed energy; acquiring the market electricity price information and demand response participation rate through an external power market data interface; acquiring the electricity price adjustment parameters through an external power generation enterprise decision system, and acquiring the safety constraint parameters through an external power grid dispatching center interface; acquiring the demand response parameters through a user-side energy management system.

3. The method of claim 2, wherein, The step of constructing the behavior influence matrix according to the real-time state and multi-agent decision data comprises: extracting price fluctuation features from the market electricity price information, and extracting load response features from the demand response participation rate, to form a market fluctuation feature set comprising the price fluctuation features and the load response features; calculating the income influence weight of each agent decision variable relative to other agents based on the market fluctuation feature set, the electricity price adjustment parameters, the demand response parameters and the safety constraint parameters; constructing an initial behavior influence matrix according to each income influence weight, wherein each element in the initial behavior influence matrix is used to represent a correlation coefficient between decision variables; performing correlation analysis on the environmental parameters and the real-time total power output of the virtual power plant by using a time series analysis method, to update dynamic coupling parameters in the initial behavior influence matrix; performing feasibility check on the updated initial behavior influence matrix through the operating state data, and outputting the behavior influence matrix after the feasibility check is completed.

4. The method of claim 3, wherein, The step of constructing the hybrid action space model based on the behavior influence matrix comprises: extracting coupling relationships and conflict weights between multi-agent decision variables from the behavior influence matrix, to determine the spatial boundary constraints of the discrete actions and the continuous actions; defining device start-stop states and switch switching instructions as discrete action spaces, and defining power distribution ratios and output adjustment amounts as continuous action spaces; The space boundary constraint input is input into a preset neural network model, joint representation and dimension fusion operations are performed on the discrete action space and the continuous action space, and a hybrid action representation vector is output; A state-action mapping rule set is constructed according to the hybrid action representation vector and dynamic coupling parameters in the behavior influence matrix; An initial scheduling policy set is generated by using a policy gradient algorithm based on the state-action mapping rule set; Action conflict resolution and feasibility verification are performed on the initial scheduling policy set, and a hybrid action space model is constructed according to the initial scheduling policy set after action conflict resolution and feasibility verification.

5. The method of claim 4, wherein, Based on the unified scheduling policy framework, an evolutionary algorithm is used to optimize the discrete action and the continuous action to obtain a multi-region collaborative path, including: Power balance constraints, transmission capacity constraints, and response time constraints are extracted from the unified scheduling policy framework to form a constraint condition set; An initial population is generated according to the constraint condition set, the initial population including multiple candidate collaborative paths, each candidate collaborative path including a space sequence of corresponding discrete actions and parameterized instructions of continuous actions; Based on the benefit influence weight and conflict weight in the behavior influence matrix, and combined with the constraint condition set, the comprehensive fitness of each candidate collaborative path is calculated; Selection operation, crossover operation, and mutation operation are used to iteratively evolve the initial population, the crossover operation including fragment recombination of discrete action sequences and arithmetic crossover of continuous action parameters, and the mutation operation including random flipping of discrete actions and Gaussian disturbance of continuous actions; When the maximum number of iterations is reached, or the comprehensive fitness of each candidate collaborative path converges to a set threshold, the candidate collaborative path with the highest comprehensive fitness is selected from the current population as the optimized multi-region collaborative path.

6. The method of claim 5, wherein, Based on the multi-region collaborative path, resource aggregation and power allocation attributes are fused to construct a service interaction network, including: Cross-region power exchange plans and device scheduling instructions are extracted from the multi-region collaborative path to generate an initial resource allocation scheme; Based on resource aggregation characteristics, the initial resource allocation scheme is subjected to schedulable capacity verification, the resource aggregation characteristics including response rate, adjustable capacity range, and minimum residence time of distributed resources; According to the set power allocation attributes, the verified resource allocation scheme is subjected to regional power balance optimization, the power allocation attributes including transmission loss coefficient, node voltage limit, and power flow safety constraint; Based on the benefit influence weight in the behavior influence matrix, the optimized power allocation scheme is subjected to benefit conflict detection and balance adjustment; A multi-layer network topology structure is created, including a physical layer, an information layer, and a service layer; The adjusted resource allocation scheme is mapped to the physical layer, the optimized power allocation scheme is mapped to the information layer, and a state monitoring module and a protocol adaptation module are integrated in the service layer of the multi-layer network topology structure to generate the service interaction network.

7. The method of claim 6, wherein, The coordination optimization sequence is mapped to the service interaction network to output a hierarchical partition coordination model, including: extracting a cross-layer benefit adjustment parameter and a partition coordination agreement from the coordination optimization sequence, generating a hierarchical coordination rule set; based on the state monitoring and protocol adaptation results in the business interaction network, dynamically correcting the hierarchical coordination rule set to obtain an updated hierarchical coordination rule; according to the updated hierarchical coordination rule, reconstructing a partition topology structure of the virtual power plant, the partition topology structure including a partition boundary adjustment mechanism and a cross-layer power balance constraint; mapping the reconstructed partition topology structure to the physical layer and the information layer of the business interaction network, generating a partition coordination basic framework; based on the partition coordination basic framework and the updated hierarchical coordination rule, constructing a hierarchical decision engine; outputting the hierarchical partition coordination model by executing the coordination optimization sequence through the hierarchical decision engine.

8. A hierarchical zonal modeling system for a virtual power plant, characterized by, comprising: a behavior influence matrix construction module for obtaining real-time state and multi-agent decision data of distributed energy in a virtual power plant, and constructing a behavior influence matrix according to the real-time state and multi-agent decision data; a scheduling strategy framework construction module for constructing a hybrid action space model based on the behavior influence matrix, and generating a unified scheduling strategy framework through the hybrid action space model, the unified scheduling strategy framework including discrete actions and continuous actions; a regional coordination path generation module for optimizing the discrete actions and continuous actions based on the unified scheduling strategy framework using an evolutionary algorithm to obtain a multi-region coordination path; a coordination optimization sequence generation module for constructing a business interaction network based on the multi-region coordination path by fusing resource aggregation and power distribution attributes, and generating a coordination optimization sequence for driving the business interaction network to run; a hierarchical partition model construction module for mapping the coordination optimization sequence to the business interaction network to output a hierarchical partition coordination model.

9. An electronic device, comprising: comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, Computer instructions for causing a computer to perform the method according to any one of claims 1-7.

Images