Power grid equipment time-phased power failure strategy self-adaptive generation system and method based on GRPO algorithm

Abstract

The invention discloses a GRPO algorithm-based power grid equipment time-phased power failure strategy adaptive generation system, which is characterized in that an agent set, a state space, an action space, a reward function and a state transfer function are defined based on a Markov decision process, and a multi-agent model is established in combination with an objective function; respectively constructing constraint conditions and forming a constraint condition set through a security-guaranteed target, a supply-guaranteed target and a consumption-guaranteed target of power grid operation; based on the constraint condition set, using a group relative strategy optimization algorithm to carry out iterative optimization training on the multi-agent model, and generating a trained multi-agent model; and dynamically adjusting power failure time, power failure duration and maintenance resource configuration of power grid equipment based on actual power grid data through the trained multi-agent model, and generating an optimal time-phased power failure strategy. According to the invention, efficient and adaptive generation of the power grid power failure strategy is realized, the arrangement efficiency is improved, and the operation risk is reduced.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to an adaptive generation system and method for time-sharing outage strategies of power grid equipment based on the GRPO algorithm. Background Technology

[0002] With the increasing complexity of power system operation characteristics and control modes, power grid operation is becoming more and more varied, and the number of objects under its dispatch and control is also growing significantly. The number of devices in the power grid is increasing, and the proportion of new energy sources such as wind power and photovoltaics connected to the grid is gradually rising. The uncertainty, volatility, and indirectness of new energy output pose serious challenges to the safe and reliable operation of the power grid, especially impacting the scheduling of power outage strategies. Ensuring power grid operation safety, power supply, and new energy consumption during power outages is a critical challenge that existing solutions urgently need to overcome. Heuristic algorithms based on swarm intelligence optimization, such as particle swarm optimization, dragonfly swarm optimization, and ant colony optimization, can only avoid local optima to a certain extent, but cannot adapt to the constantly changing power grid environment in real time. Furthermore, as the number of power grid devices increases, the power outage strategy scheduling model becomes more complex, leading to a heavier computational burden on traditional heuristic algorithms, resulting in long iterative solution times, slow convergence speeds, and low computational efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive generation system and method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm. This invention achieves efficient and adaptive generation of power grid outage strategies, improves scheduling efficiency, and reduces operational risks.

[0004] To achieve this objective, the present invention provides an adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, comprising: The model building module is used to define the set of agents, state space, action space, reward function and state transition function based on Markov decision process, and to build a multi-agent model in combination with the objective function; The constraint integration module is used to construct constraint conditions and form a constraint condition set based on the power grid operation's safety, supply, and consumption targets. The model training module is used to iteratively optimize and train the multi-agent model based on a set of constraints using a group relative policy optimization algorithm, thereby generating the trained multi-agent model. The strategy generation module is used to dynamically adjust the outage time, outage duration, and maintenance resource allocation of power grid equipment based on actual power grid data using the trained multi-agent model, and generate the optimal time-sharing outage strategy.

[0005] Preferably, in the model building module, the definition of the agent set, state space, action space, reward function, and state transition function based on Markov decision processes is specifically used for: Based on Markov decision processes, tuples are defined. ; in, A collection of intelligent agents, For state space, For the action space, For the reward function, This is the state transition function. is the discount factor for the objective function; Collection of intelligent agents This includes intelligent agents for generator sets, intelligent agents for power lines, and intelligent agents for transformers. state space The expression is: in, For the load gap, Contribute to new energy sources To score risk, For maintenance plan matrix, For resource events, For special events, For the health status of power grid equipment, Power outage sequence for users; The load gap ;in, The proportion of the load gap in the entire province. ; The proportion of the load gap in prefecture-level cities. The proportion of load gap in county-level cities; The new energy output The ratio of renewable energy generation to total grid generation is expressed as: in, To contribute to wind power, Contribute to photovoltaic power The total power generation of the power grid; The risk score The risk scores, including safety risk score, supply risk score, and absorption risk score, are all calculated using the risk scoring formula, expressed as follows: in, This indicates the weight of the corresponding risk event. This indicates the number of such risk events. Indicates the number of days the risk event lasted; The maintenance plan matrix The expression is: in, For power grid equipment i In the t The maintenance status of the day; The resource event ;in This represents the number of currently available resources for the generator set. This represents the number of currently available resources for the line. This represents the number of currently available resources for the transformer. The special event ;in Indicates the first t The value is 1 when the event is a special event, otherwise it is 0. The health status of the power grid equipment ;in For power grid equipment j Health index; The user power outage sequence ;in For users i In the t The power outage lasted for 1 day; Action space Represented as : Time travel ;in Indicates power grid equipment i The scheduled power outage for maintenance has been moved. For equipment i Maintenance time offset; Duration adjustment Among them are power grid equipment. i The power outage duration adjustment amount; Resource scheduling ;in Adjustments to maintenance resources for generator sets. Adjustments to the maintenance resources for the line. Adjustments to transformer maintenance resources; reward function It consists of risk reduction rewards, action cost penalties, and constraint violation penalties, expressed as follows: for: in, As a weighting factor for safety risk scoring, The change in the safety risk score. For supply risk scoring weights, This represents the change in the supply risk score. To absorb the risk scoring weights, To absorb changes in risk scores, To constrain the penalty coefficient for violations, To constrain the identification of violations, These are the weighting coefficients for the health status of power grid equipment and user satisfaction. Indicates the health status of power grid equipment. This indicates the user's level of satisfaction with the power grid's operating status; State transition matrix Represented as : in, This indicates the current state of the power grid. Indicates the state of the power grid at the next moment. This is the maintenance plan matrix after the action update. This is a power flow calculation model.

[0006] Preferably, in the model building module, the specific method for establishing a multi-agent model by combining the objective function is as follows: The expression for the objective function J is: in, For discount factors, 0 < <1; Let represent the expectation operator, and let represent the optimization of minimizing policy π. As a weighting factor for safety risk scoring, for t Safety score at all times For supply risk scoring weights, for t Supply score at any moment To absorb the risk scoring weights, for t The score for absorption at any given moment; The initial state of the power grid is input, and the decision-making strategies of the agents are initialized. The current state of the power grid is input to each agent. Each agent selects an action from the action space based on its corresponding power grid equipment attributes and the current state of the power grid. The actions of all agents are executed, the maintenance plan matrix is ​​updated, and the new state of the power grid is calculated through the power flow simulation model. Based on the difference between the new state of the power grid and the current state of the power grid, the reward value is calculated through the reward function and fed back to all agents. The agents adjust their decision-making strategies according to the reward value. The above process is iterated until the objective function converges, and the multi-agent model is completed.

[0007] Preferably, in the constraint integration module, the constraint conditions are constructed and a constraint condition set is formed by the power grid operation's safety assurance target, supply assurance target, and consumption assurance target, respectively, specifically for: The safety objectives of power grid operation are used to constrain the operational boundaries of power grid equipment, including: Power flow constraints are used to limit the power flow values ​​of a line: ,in, For a moment t line l Trend value, For the line l Permissible power flow limits; Node voltage constraints are used to limit the voltage value at each circuit node in a power grid: ,in, For a moment t line l voltage value, For the circuit nodes of the power grid equipment at time t line l The lower limit of voltage, For the circuit nodes of the power grid equipment at time t line l The upper limit of voltage; The power supply guarantee objective of power grid operation is used to constrain the power supply reliability of power grid equipment, minimizing the number and duration of power outages for users. For power grid equipment whose maintenance would cause repeated power outages to the same area, the corresponding power grid equipment will be scheduled for maintenance within the same time period. ,in, Indicates the first i Planned maintenance time for each piece of equipment. Indicates the first j Planned maintenance time for each piece of equipment; The grid operation's consumption guarantee targets are used to constrain the consumption of new energy sources and the flexibility of the grid system, including: Mutual exclusion constraints are used to stagger the maintenance times of grid equipment that would weaken the grid system or cause additional power outages during maintenance. ; The curtailment constraint is used to ensure that at least one connecting line of a renewable energy unit is in normal operation. If all outgoing lines are scheduled to be out of power, the renewable energy unit will be forced to curtail power. The constraint is defined as the number of non-outage states of the relevant lines being greater than zero. ,in, This represents the total number of lines in a non-power-outage state among all the connection lines of the new energy generating unit; Finally, all the above constraints are combined into a constraint set.

[0008] Preferably, in the model training module, based on the constraint set, the multi-agent model is iteratively optimized and trained using a group relative policy optimization algorithm to generate a trained multi-agent model, specifically used for: Initialize the parameters of the multi-agent model and embed the complete constraint rules constructed based on the constraint condition set into the power grid power flow simulation environment; The GRPO algorithm is used to perform multiple iterations on the multi-agent model. For each iteration: A batch of power grid states are randomly selected, and G complete power outage plan sequences are generated for each power grid state using a fixed old strategy. Each generated power outage plan sequence is input into the power grid power flow simulation environment. The power grid power flow simulation environment calls the constraint condition set to verify whether each power outage plan violates any constraints, and simulates and deduces the changes in the power grid state after executing the power outage plan. Based on the simulation results and the reward function, the total reward for each power outage plan sequence is calculated. The GRPO objective function is calculated based on the data in the defined tuple, and its expression is: in, This represents the objective function that needs to be maximized. The parameters represent the multi-agent model. It represents the mathematical expectation over all possible grid states and possible outage plans; q To be from the distribution One of the prompts in the sampling, namely the power grid status, represents a snapshot of the current power grid status; Represents the old strategy model; This indicates that the old multi-agent model is based on prompts. q A set of G completions is generated, representing G distinct candidate power outage plans; Indicates the first i A sequence of power outage plans; O To complete the sequence of power outage plans; This represents the average value of each completed set. This represents the average value of each set of power outage plans; Indicates the first i The power outage plan sequence is in the first t The action selected at each time step; This indicates that the old strategy model generates actions. The probability of; This indicates that the new policy model generates actions. The probability of; This indicates that the old and new strategies apply to the same action. The proportion of change in preferences; Indicates the relative advantage estimate of the group; The function is used to limit the value between arrive between; The regularization coefficient is used. KL divergence is used to measure the new strategy. Reference Strategy The differences between them; New computational strategies Reference Strategy The KL divergence between them is used as a penalty term, and its expression is: in, Indicates reference strategy Generate Actions The probability of; By continuously adjusting the parameters of the policy model to maximize the GRPO objective function, the training loop terminates when the preset maximum number of iterations is reached, and the final policy model is the trained multi-agent model.

[0009] Preferably, in the strategy generation module, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment based on actual power grid data, thereby generating an optimal time-sharing outage strategy, specifically used for: Based on actual power grid data, and according to the state space defined in the training process. The process involves cleaning, integrating, and standardizing the data to construct state vectors. These state vectors are then input into a trained multi-agent model, which directly outputs the optimal action probability distribution for the current state. Based on this optimal action probability distribution, the power grid system selects the action space with the highest probability. According to the action space The outage time, duration, and maintenance resource allocation of power grid equipment are dynamically adjusted to generate a preliminary outage strategy. This preliminary outage strategy is then quickly verified in a power grid power flow simulation environment with centralized constraints. If the verification passes, the optimal time-sharing outage strategy is obtained.

[0010] This invention also provides an adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, comprising: Based on Markov decision processes, we define the set of agents, state space, action space, reward function, and state transition function, and establish a multi-agent model by combining the objective function. Constraints are constructed and a set of constraints is formed by considering the power grid operation's objectives of ensuring safety, ensuring supply, and ensuring consumption. Based on the set of constraints, the multi-agent model is iteratively optimized and trained using a group relative policy optimization algorithm to generate a trained multi-agent model. Based on actual power grid data, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment to generate the optimal time-sharing outage strategy.

[0011] The present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm as described above.

[0012] The beneficial effects of this invention are: This invention, through the GRPO algorithm, avoids drastic fluctuations in outage strategy updates while ensuring long-term optimization stability and improving the efficiency of outage strategy orchestration. By dynamically adjusting the outage time, duration, and maintenance resource allocation of grid equipment, and combining multiple constraints of ensuring safety, supply, and consumption, it significantly reduces the safety, supply, and consumption risks of grid operation. Furthermore, by directly using the reward function to guide outage strategy updates through the GRPO algorithm, it can respond to changes in grid status in real time, adapting to complex and ever-changing grid environments. Simultaneously, through multi-agent collaborative decision-making and rigorous constraint verification, it ensures the safety of grid operation, the reliability of power supply, and the effectiveness of renewable energy consumption, thereby improving the health status of grid equipment and user satisfaction.

[0013] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0014] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0015] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0016] Example 1 An adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, such as... Figure 1 As shown, it includes: The model building module is used to define a set of agents (each power grid device is defined as an agent), state space, action space, reward function, and state transition function based on Markov decision process, and to build a multi-agent model in combination with objective function. This design provides a quantitative basis for agent decision optimization by defining the set of agents, state space, action space, reward function, and state transition function. The multi-agent model built in combination with objective function can comprehensively capture the complex characteristics and decision requirements of power grid operation. The constraint integration module is used to construct constraints and form a constraint set based on the power grid operation's safety, supply, and consumption objectives. This design provides rigid specifications for the agent's decision-making through the constraint set, ensuring that the generated power outage strategy is feasible and safe. The model training module is used to iteratively optimize and train the multi-agent model based on the constraint set using the Group Relative Policy Optimization (GRPO) algorithm, generating a trained multi-agent model. This design uses the GRPO algorithm to iteratively optimize and train the multi-agent model, guiding the agents to optimize in the direction of reducing risks and improving efficiency, while avoiding drastic fluctuations in policy updates and forgetting historical optimal policies. At the same time, the constraint rules of the constraint set are embedded into the power grid power flow simulation environment to verify the compliance of decisions in real time during the training process, ensuring that the trained multi-agent model has good decision-making performance and stability. The strategy generation module is used to dynamically adjust the outage time, duration, and maintenance resource allocation of power grid equipment based on actual power grid data using the trained multi-agent model, generating the optimal time-sharing outage strategy. This design utilizes the trained multi-agent model, combined with actual power grid data to construct a state vector, and dynamically adjusts the outage time, duration, and maintenance resource allocation based on the optimal action probability distribution output by the trained multi-agent model. After generating the preliminary strategy, it is quickly verified through a power grid power flow simulation environment and constraint set to ensure that the strategy conforms to the actual power grid operating state and meets all constraint requirements. Finally, the optimal time-sharing outage strategy is output, ensuring the adaptability, accuracy, and feasibility of the outage strategy.

[0017] In the above technical solution, the definition of the agent set, state space, action space, reward function, and state transition function based on Markov decision process is specifically used for: Based on Markov decision processes, tuples are defined. ; in, A collection of intelligent agents, For state space, For the action space, For the reward function, This is the state transition function. is the discount factor for the objective function; Collection of intelligent agents This includes intelligent agents for generator sets, intelligent agents for power lines, and intelligent agents for transformers. state space The expression is: in, For the load gap, Contribute to new energy sources To score risk, For maintenance plan matrix, For resource events, For special events, For the health status of power grid equipment, Power outage sequence for users; The load gap ;in, The proportion of the load gap in the entire province. ; The proportion of the load gap in prefecture-level cities. The proportion of load gap in county-level cities; The new energy output The ratio of renewable energy generation to total grid generation is expressed as: in, To contribute to wind power, Contribute to photovoltaic power The total power generation of the power grid; The risk score The risk scores, including safety risk score, supply risk score, and absorption risk score, are all calculated using the risk scoring formula, expressed as follows: in, This indicates the weight of the corresponding risk event. This indicates the number of such risk events. Indicates the number of days the risk event lasted; The maintenance plan matrix The expression is: in, For power grid equipment i In the t The maintenance status of the day; The resource event ;in This represents the number of currently available resources for the generator set. This represents the number of currently available resources for the line. This represents the number of currently available resources for the transformer. The special event ;in Indicates the first t The value is 1 when the event is a special event, otherwise it is 0. The health status of the power grid equipment ;in For power grid equipment j Health index; The user power outage sequence ;in For users i In the t The power outage lasted for 1 day; Action space Represented as : Time travel ;in Indicates power grid equipment i The scheduled power outage for maintenance has been moved. For equipment i Maintenance time offset; Duration adjustment Among them are power grid equipment. i The power outage duration adjustment amount; Resource scheduling ;in Adjustments to maintenance resources for generator sets. Adjustments to the maintenance resources for the line. Adjustments to transformer maintenance resources; reward function It consists of risk reduction rewards, action cost penalties, and constraint violation penalties, expressed as follows: for: in, As a weight for the safety risk score (can take a value of 0.4). The change in the safety risk score. The weight for supply risk scoring (can be 0.4). This represents the change in the supply risk score. To absorb the risk score weight (which can be 0.2). To absorb changes in risk scores, To constrain the penalty coefficient for violations (violations of human resources or mutually exclusive relationships), (Value can be 100) To constrain the identification of violations, These are the weighting coefficients for the health status of power grid equipment and user satisfaction. Indicates the health status of power grid equipment. This indicates the user's level of satisfaction with the power grid's operating status; State transition matrix Represented as : in, This indicates the current state of the power grid. Indicates the state of the power grid at the next moment. This is the maintenance plan matrix after the action update. The above design defines the components of the Markov decision process tuple, providing a clear quantitative standard for the decision-making process of the multi-agent model. This enables the multi-agent model to more accurately capture changes in the power grid state, quantify the decision-making effect, and improve the scientific nature and operability of the multi-agent model's decision-making.

[0018] In the above technical solution, the specific method for establishing a multi-agent model by combining the objective function is as follows: The expression for the objective function J is: in, For discount factors, 0 < <1; represents the expectation operator (used to calculate the expected value of a random variable), and represents the optimization of minimizing the policy π. As a weighting factor for safety risk scoring, for t Safety score at all times For supply risk scoring weights, for t Supply score at any moment To absorb the risk scoring weights, for t The score for absorption at any given moment; The process involves inputting the initial state of the power grid, initializing the decision-making strategies of the agents, inputting the current power grid state to each agent, and having each agent select an action from the action space based on its corresponding power grid equipment attributes and the current power grid state. The actions of all agents are then executed, the maintenance plan matrix is ​​updated, and the new power grid state is calculated using a power flow simulation model. Based on the difference between the new power grid state and the current power grid state, a reward value is calculated using a reward function and fed back to all agents. Agents adjust their decision-making strategies based on the reward value, and the process iterates until the objective function converges, thus completing the construction of the multi-agent model. This design clarifies the construction logic and optimization direction of the multi-agent model, ensuring that all agents can make collaborative decisions based on a common goal, avoiding the one-sidedness of individual agent decisions.

[0019] In the above technical solution, the constraint conditions are constructed and a constraint condition set is formed by using the power grid operation's safety, supply, and consumption targets respectively, specifically for: The safety objectives of power grid operation are used to constrain the operational boundaries of power grid equipment, including: Power flow constraints are used to limit the power flow values ​​of a line: ,in, For a moment t line l Trend value, For the line l Permissible power flow limits; Node voltage constraints are used to limit the voltage value at each circuit node in a power grid: ,in, For a moment t line l voltage value, For the circuit nodes of the power grid equipment at time t line l The lower limit of voltage, For the circuit nodes of the power grid equipment at time t line l The upper limit of voltage; The power supply guarantee objective of power grid operation is used to constrain the power supply reliability of power grid equipment, minimizing the number and duration of power outages for users. For power grid equipment whose maintenance would cause repeated power outages to the same area, the corresponding power grid equipment will be scheduled for maintenance within the same time period. ,in, Indicates the first i Planned maintenance time for each piece of equipment. Indicates the first j Planned maintenance time for each piece of equipment; The grid operation's consumption guarantee targets are used to constrain the consumption of new energy sources and the flexibility of the grid system, including: Mutual exclusion constraints are used to stagger the maintenance times of grid equipment whose maintenance would weaken the power grid system or cause additional power outages (such as two main transformers in the same substation). ; The curtailment constraint is used to ensure that at least one connecting line of a renewable energy unit is in normal operation. If all outgoing lines are scheduled to be out of power, the renewable energy unit will be forced to curtail power. The constraint is defined as the number of non-outage states of the relevant lines being greater than zero. ,in, This represents the total number of lines in a non-power-outage state among all the connection lines of the new energy generating unit; Finally, all the above constraints are combined into a constraint set; the above design clarifies the constraint details and quantitative standards of various objectives of power grid operation, making the constraint set more complete and specific, and providing more accurate rule specifications for multi-agent model training and policy generation.

[0020] In the above technical solution, based on the constraint set, a group relative policy optimization algorithm is used to iteratively optimize and train the multi-agent model to generate a trained multi-agent model, specifically used for: Initialize the parameters of the multi-agent model and embed the complete constraint rules constructed based on the constraint condition set into the power grid power flow simulation environment; The GRPO algorithm is used to perform multiple iterations on the multi-agent model. For each iteration: Randomly select a batch of power grid statuses (power grid status is indicated by the prompt) q Using a fixed old strategy, G complete outage plan sequences (i.e., complete outage plan sequences) are generated for each power grid state. Each generated outage plan sequence is input into the power grid power flow simulation environment. The power grid power flow simulation environment calls the constraint condition set to check whether each outage plan violates any constraints and simulates the changes in the power grid state after executing the outage plan. Based on the simulation results and the reward function, the total reward for each outage plan sequence is calculated. If the outage plan violates the constraints, the penalty term in the reward function will be activated, resulting in a significant reduction in the total reward, thus giving the decision model a strong negative signal. The GRPO objective function is calculated based on the data in the defined tuple, and its expression is: in, This represents the objective function that needs to be maximized (a comprehensive score used to measure the quality of the current agent model (policy model). The higher the score, the better the power outage plan generated by the policy model performs in terms of reducing risk and ensuring power supply). The parameters represent the multi-agent model. It represents the mathematical expectation for all possible grid states and possible outage plans (the goal is to make the trained strategy universal and able to cope with various different grid conditions). q To be from the distribution One of the prompts in the sampling, namely the power grid status, represents a snapshot of the current power grid status; This represents the old policy model (i.e., the old multi-agent model). This indicates that the old multi-agent model is based on prompts. q A set of G completions is generated, representing G distinct candidate power outage plans; Indicates the first i Each power outage plan sequence (each power outage plan sequence specifies in detail when, which power grid equipment will be affected, and for how long the power outage maintenance will be carried out); O To complete the sequence of power outage plans; This represents the average value of each completed set. This represents the average value of each set of power outage plans; Indicates the first i The power outage plan sequence is in the first t The action selected at each time step; This indicates that the old strategy model generates actions. The probability of; This indicates that the new policy model generates actions. The probability of; This indicates that the old and new strategies apply to the same action. The ratio of the change in preference (if the ratio is greater than 1, it means the new strategy is more inclined to make that decision; if it is less than 1, it means the new strategy is inclined to avoid that decision). Relative advantage estimate of the group (evaluation of the first group) i The power outage plan sequence is in the first t (Advantages of each time step) The function is used to limit the value between arrive between; Regularization coefficient ( This is a hyperparameter greater than 0, used to control the weight of the next KL divergence in the GRPO objective function. The larger the value, the more desired the new policy is to avoid deviating from the reference policy. KL divergence is used to measure the new strategy. Reference Strategy The difference between them (the role of KL divergence is to prevent the new strategy from becoming completely different from the old strategy, ensuring that the optimized new strategy will not make some extreme, high-risk, and unverified power outage arrangements in pursuit of high scores, thus guaranteeing the conservatism and security of the strategy). New computational strategies Reference Strategy The KL divergence between them is used as a penalty term, and its expression is: in, Indicates reference strategy Generate Actions The probability of; By continuously adjusting the parameters of the policy model to maximize the GRPO objective function, the training loop terminates when the preset maximum number of iterations is reached (the maximum number of iterations is adjusted according to the specific power grid environment, computing resources, and accuracy requirements in practical applications). The final policy model obtained is the trained multi-agent model. The above design continuously optimizes the policy model parameters through the GRPO algorithm during iteration, ensuring that the objective function value is maximized while satisfying the constraint set. At the same time, the KL divergence penalty term avoids excessive deviation between the policy and the reference model, ensuring the stability of the training process and the decision reliability of the final policy model.

[0021] In the above technical solution, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment based on actual power grid data, thereby generating an optimal time-sharing outage strategy, specifically used for: Based on actual power grid data, and according to the state space defined in the training process. The process involves cleaning, integrating, and standardizing the data to construct a standardized state vector that comprehensively describes the current power grid state. This generated state vector is then input into a trained multi-agent model, which directly outputs the optimal action probability distribution for the current state. Based on this optimal action probability distribution, the power grid system selects the action space with the highest probability. (Contains specific instructions for all power grid equipment that needs adjustment), based on the action space. The system dynamically adjusts the outage time, duration, and maintenance resource allocation for power grid equipment to generate a preliminary outage strategy. This preliminary strategy is then rapidly validated in a power grid power flow simulation environment with centralized constraints. Once the validation passes, the optimal time-sharing outage strategy is obtained. This design clarifies the generation process of the optimal time-sharing outage strategy, ensuring its accuracy, efficiency, and compliance. This allows the outage strategy to quickly adapt to actual dynamic changes in the power grid, improving the efficiency and practicality of outage strategy generation.

[0022] Example 2 like Figure 2 As shown, this embodiment provides an adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, including the following steps: S1. Based on Markov decision processes, define the set of agents, state space, action space, reward function, and state transition function. Combine this with the objective function to establish a multi-agent model, specifically including: 1.1 Based on Markov decision processes, we define the set of agents, state space, action space, reward function, and state transition function. The specific process is as follows: Based on Markov decision processes, tuples are defined. ; in, A collection of intelligent agents, For state space, For the action space, For the reward function, This is the state transition function. is the discount factor for the objective function; Collection of intelligent agents This includes intelligent agents for generator sets, intelligent agents for power lines, and intelligent agents for transformers. state space The expression is: in, For the load gap, Contribute to new energy sources To score risk, For maintenance plan matrix, For resource events, For special events, For the health status of power grid equipment, Power outage sequence for users; The load gap ;in, The proportion of the load gap in the entire province. ; The proportion of the load gap in prefecture-level cities. The proportion of load gap in county-level cities; The new energy output The ratio of renewable energy generation to total grid generation is expressed as: in, To contribute to wind power, Contribute to photovoltaic power The total power generation of the power grid; The risk score The risk scores, including safety risk score, supply risk score, and absorption risk score, are all calculated using the risk scoring formula, expressed as follows: in, This indicates the weight of the corresponding risk event. This indicates the number of such risk events. Indicates the number of days the risk event lasted; The maintenance plan matrix The expression is: in, For power grid equipment i In the t The maintenance status of the day; The resource event ;in This represents the number of currently available resources for the generator set. This represents the number of currently available resources for the line. This represents the number of currently available resources for the transformer. The special event ;in Indicates the first t The value is 1 when the event is a special event, otherwise it is 0. The health status of the power grid equipment ;in For power grid equipment j Health index; The user power outage sequence ;in For users i In the t The power outage lasted for 1 day; Action space Represented as : Time travel ;in Indicates power grid equipment i The scheduled power outage for maintenance has been moved. For equipment i Maintenance time offset; Duration adjustment Among them are power grid equipment. i The power outage duration adjustment amount; Resource scheduling ;in Adjustments to maintenance resources for generator sets. Adjustments to the maintenance resources for the line. Adjustments to transformer maintenance resources; reward function It consists of risk reduction rewards, action cost penalties, and constraint violation penalties, expressed as follows: for: in, As a weighting factor for safety risk scoring, The change in the safety risk score. For supply risk scoring weights, This represents the change in the supply risk score. To absorb the risk scoring weights, To absorb changes in risk scores, To constrain the penalty coefficient for violations, To constrain the identification of violations, These are the weighting coefficients for the health status of power grid equipment and user satisfaction. Indicates the health status of power grid equipment. This indicates the user's level of satisfaction with the power grid's operating status; State transition matrix Represented as : in, This indicates the current state of the power grid. Indicates the state of the power grid at the next moment. This is the maintenance plan matrix after the action update. This is a power flow calculation model.

[0023] 1.2 A multi-agent model is established by combining the objective function. The specific process is as follows: The expression for the objective function J is: in, For discount factors, 0 < <1; Let represent the expectation operator, and let represent the optimization of minimizing policy π. As a weighting factor for safety risk scoring, for t Safety score at all times For supply risk scoring weights, for t Supply score at any moment To absorb the risk scoring weights, for t The score for absorption at any given moment; The initial state of the power grid is input, and the decision-making strategies of the agents are initialized. The current state of the power grid is input to each agent. Each agent selects an action from the action space based on its corresponding power grid equipment attributes and the current state of the power grid. The actions of all agents are executed, the maintenance plan matrix is ​​updated, and the new state of the power grid is calculated through the power flow simulation model. Based on the difference between the new state of the power grid and the current state of the power grid, the reward value is calculated through the reward function and fed back to all agents. The agents adjust their decision-making strategies according to the reward value. The above process is iterated until the objective function converges, and the multi-agent model is completed.

[0024] S2. Constraints are constructed and a set of constraints is formed based on the power grid operation's goals of ensuring safety, supply, and consumption. Specifically, these include: The safety objectives of power grid operation are used to constrain the operational boundaries of power grid equipment, including: Power flow constraints are used to limit the power flow values ​​of a line: ,in, For a moment t line l Trend value, For the line l Permissible power flow limits; Node voltage constraints are used to limit the voltage value at each circuit node in a power grid: ,in, For a moment t line l voltage value, For the circuit nodes of the power grid equipment at time t line l The lower limit of voltage, For the circuit nodes of the power grid equipment at time t line l The upper limit of voltage; The power supply guarantee objective of power grid operation is used to constrain the power supply reliability of power grid equipment, minimizing the number and duration of power outages for users. For power grid equipment whose maintenance would cause repeated power outages to the same area, the corresponding power grid equipment will be scheduled for maintenance within the same time period. ,in, Indicates the first i Planned maintenance time for each piece of equipment. Indicates the first j Planned maintenance time for each piece of equipment; The grid operation's consumption guarantee targets are used to constrain the consumption of new energy sources and the flexibility of the grid system, including: Mutual exclusion constraints are used to stagger the maintenance times of grid equipment that would weaken the grid system or cause additional power outages during maintenance. ; The curtailment constraint is used to ensure that at least one connecting line of a renewable energy unit is in normal operation. If all outgoing lines are scheduled to be out of power, the renewable energy unit will be forced to curtail power. The constraint is defined as the number of non-outage states of the relevant lines being greater than zero. ,in, This represents the total number of lines in a non-power-outage state among all the connection lines of the new energy generating unit; Finally, all the above constraints are combined into a constraint set.

[0025] S3. Based on the constraint set, the multi-agent model is iteratively optimized and trained using a group relative policy optimization algorithm to generate a trained multi-agent model, specifically including: Initialize the parameters of the multi-agent model and embed the complete constraint rules constructed based on the constraint condition set into the power grid power flow simulation environment; The GRPO algorithm is used to perform multiple iterations on the multi-agent model. For each iteration: A batch of power grid states are randomly selected, and G complete power outage plan sequences are generated for each power grid state using a fixed old strategy. Each generated power outage plan sequence is input into the power grid power flow simulation environment. The power grid power flow simulation environment calls the constraint condition set to verify whether each power outage plan violates any constraints, and simulates and deduces the changes in the power grid state after executing the power outage plan. Based on the simulation results and the reward function, the total reward for each power outage plan sequence is calculated. The GRPO objective function is calculated based on the data in the defined tuple, and its expression is: in, This represents the objective function that needs to be maximized. The parameters represent the multi-agent model. It represents the mathematical expectation over all possible grid states and possible outage plans; q To be from the distribution One of the prompts in the sampling, namely the power grid status, represents a snapshot of the current power grid status; Represents the old strategy model; This indicates that the old multi-agent model is based on prompts. q A set of G completions is generated, representing G distinct candidate power outage plans; Indicates the first i A sequence of power outage plans; O To complete the sequence of power outage plans; This represents the average value of each completed set. This represents the average value of each set of power outage plans; Indicates the firsti The power outage plan sequence is in the first t The action selected at each time step; This indicates that the old strategy model generates actions. The probability of; This indicates that the new policy model generates actions. The probability of; This indicates that the old and new strategies apply to the same action. The proportion of change in preferences; Indicates the relative advantage estimate of the group; The function is used to limit the value between arrive between; The regularization coefficient is used. KL divergence is used to measure the new strategy. Reference Strategy The differences between them; New computational strategies Reference Strategy The KL divergence between them is used as a penalty term, and its expression is: in, Indicates reference strategy Generate Actions The probability of; By continuously adjusting the parameters of the policy model to maximize the GRPO objective function, the training loop terminates when the preset maximum number of iterations is reached, and the final policy model is the trained multi-agent model.

[0026] S4. Based on actual power grid data, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment to generate an optimal time-sharing outage strategy, specifically including: Based on actual power grid data, and according to the state space defined in the training process. The process involves cleaning, integrating, and standardizing the data to construct state vectors. These state vectors are then input into a trained multi-agent model, which directly outputs the optimal action probability distribution for the current state. Based on this optimal action probability distribution, the power grid system selects the action space with the highest probability. According to the action space The outage time, duration, and maintenance resource allocation of power grid equipment are dynamically adjusted to generate a preliminary outage strategy. This preliminary outage strategy is then quickly verified in a power grid power flow simulation environment with centralized constraints. If the verification passes, the optimal time-sharing outage strategy is obtained.

[0027] Example 3 This embodiment provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in Embodiment 2.

[0028] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0029] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will 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 apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0030] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0031] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable 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.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

[0033] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. An adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, characterized in that, It includes: The model building module is used to define the set of agents, state space, action space, reward function and state transition function based on Markov decision process, and to build a multi-agent model in combination with the objective function; The constraint integration module is used to construct constraint conditions and form a constraint condition set based on the power grid operation's safety, supply, and consumption targets. The model training module is used to iteratively optimize and train the multi-agent model based on a set of constraints using a group relative policy optimization algorithm, thereby generating the trained multi-agent model. The strategy generation module is used to dynamically adjust the outage time, outage duration, and maintenance resource allocation of power grid equipment based on actual power grid data using the trained multi-agent model, and generate the optimal time-sharing outage strategy.

2. The adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 1, characterized in that, In the model building module, the definition of the agent set, state space, action space, reward function, and state transition function based on Markov decision processes is specifically used for: Based on Markov decision processes, tuples are defined. ; in, For a collection of intelligent agents, For state space, For the action space, For the reward function, This is the state transition function. is the discount factor for the objective function; Collection of intelligent agents This includes intelligent agents for generator sets, intelligent agents for power lines, and intelligent agents for transformers. state space The expression is: in, For the load gap, Contribute to new energy sources To score the risk, For maintenance plan matrix, For resource events, For special events, For the health status of power grid equipment, Power outage sequence for users; The load gap ;in, The proportion of the load gap in the entire province, ; The proportion of the load gap in prefecture-level cities. The proportion of load gap in county-level cities; The new energy output The ratio of renewable energy generation to total grid generation is expressed as: in, To provide power for wind power, Contribute to photovoltaic power The total power generation of the power grid; The risk score The risk scores, including safety risk score, supply risk score, and absorption risk score, are all calculated using the risk scoring formula, expressed as follows: in, This indicates the weight of the corresponding risk event. This indicates the number of such risk events. Indicates the number of days the risk event lasted; The maintenance plan matrix The expression is: in, For power grid equipment i In the t The maintenance status of the day; The resource event ;in This represents the number of currently available resources for the generator set. This represents the number of currently available resources for the line. This represents the number of currently available resources for the transformer. The special event ;in Indicates the first t The value is 1 when the event is a special event, otherwise it is 0. The health status of the power grid equipment ;in For power grid equipment j Health index; The user power outage sequence ;in For users i In the t The power outage lasted for 1 day; Action space Represented as : Time travel ;in Indicates power grid equipment i The scheduled power outage for maintenance has been moved. For equipment i Maintenance time offset; Duration adjustment Among them are power grid equipment. i The power outage duration adjustment amount; Resource scheduling ;in Adjustments to maintenance resources for generator sets. Adjustments to the maintenance resources for the line. Adjustments to transformer maintenance resources; reward function It consists of risk reduction rewards, action cost penalties, and constraint violation penalties, expressed as follows: for: in, As a weighting factor for safety risk scoring, The change in the safety risk score. For supply risk scoring weights, This represents the change in the supply risk score. To absorb the risk scoring weights, To absorb changes in risk scores, To constrain the penalty coefficient for violations, To constrain the identification of violations, These are the weighting coefficients for the health status of power grid equipment and user satisfaction. Indicates the health status of power grid equipment. This indicates the user's level of satisfaction with the power grid's operating status; State transition matrix Represented as : in, This indicates the current state of the power grid. Indicates the state of the power grid at the next moment. This is the maintenance plan matrix after the action update. This is a power flow calculation model.

3. The adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 2, characterized in that, In the model building module, the specific method for establishing a multi-agent model by combining the objective function is as follows: The expression for the objective function J is: in, The discount factor is 0 < <1; Let represent the expectation operator, and let represent the optimization of minimizing policy π. As a weighting factor for safety risk scoring, for t Safety score at all times For supply risk scoring weights, for t Supply score at any moment To absorb the risk scoring weights, for t The score for absorption at any given moment; The initial state of the power grid is input, and the decision-making strategies of the agents are initialized. The current state of the power grid is input to each agent. Each agent selects an action from the action space based on its corresponding power grid equipment attributes and the current state of the power grid. The actions of all agents are executed, the maintenance plan matrix is ​​updated, and the new state of the power grid is calculated through the power flow simulation model. Based on the difference between the new state of the power grid and the current state of the power grid, the reward value is calculated through the reward function and fed back to all agents. The agents adjust their decision-making strategies according to the reward value. The above process is iterated until the objective function converges, and the multi-agent model is completed.

4. The adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 1, characterized in that, In the constraint integration module, the constraint conditions are constructed and a constraint condition set is formed by the power grid operation's safety assurance target, supply assurance target, and consumption assurance target, respectively. Specifically, this is used for: The safety objectives of power grid operation are used to constrain the operational boundaries of power grid equipment, including: Power flow constraints are used to limit the power flow values ​​of a line: ,in, For a moment t line l Trend value, For the line l Permissible power flow limits; Node voltage constraints are used to limit the voltage value at each circuit node in a power grid: ,in, For a moment t line l voltage value, For the circuit nodes of the power grid equipment at time t line l The lower limit of voltage, For the circuit nodes of the power grid equipment at time t line l The upper limit of voltage; The power supply guarantee objective of power grid operation is used to constrain the power supply reliability of power grid equipment, minimizing the number and duration of power outages for users. For power grid equipment whose maintenance would cause repeated power outages to the same area, the corresponding power grid equipment will be scheduled for maintenance within the same time period. ,in, Indicates the first i Planned maintenance time for each piece of equipment. Indicates the first j Planned maintenance time for each piece of equipment; The grid operation's consumption guarantee targets are used to constrain the consumption of new energy sources and the flexibility of the grid system, including: Mutual exclusion constraints are used to stagger the maintenance times of grid equipment that would weaken the grid system or cause additional power outages during maintenance. ; The curtailment constraint is used to ensure that at least one connecting line of a renewable energy unit is in normal operation. If all outgoing lines are scheduled to be out of power, the renewable energy unit will be forced to curtail power. The constraint is defined as the number of non-outage states of the relevant lines being greater than zero. ,in, This represents the total number of lines in a non-power-outage state among all the connection lines of the new energy generating unit; Finally, all the above constraints are combined into a constraint set.

5. The adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 1, characterized in that, In the model training module, based on the constraint set, the group relative policy optimization algorithm is used to iteratively optimize and train the multi-agent model, generating a trained multi-agent model, specifically used for: Initialize the parameters of the multi-agent model and embed the complete constraint rules constructed based on the constraint condition set into the power grid power flow simulation environment; The GRPO algorithm is used to perform multiple iterations on the multi-agent model. For each iteration: A batch of power grid states are randomly selected, and G complete power outage plan sequences are generated for each power grid state using a fixed old strategy. Each generated power outage plan sequence is input into the power grid power flow simulation environment. The power grid power flow simulation environment calls the constraint condition set to verify whether each power outage plan violates any constraints, and simulates and deduces the changes in the power grid state after executing the power outage plan. Based on the simulation results and the reward function, the total reward for each power outage plan sequence is calculated. The GRPO objective function is calculated based on the data in the defined tuple, and its expression is: in, This represents the objective function that needs to be maximized; The parameters represent the multi-agent model. It represents the mathematical expectation over all possible grid states and possible outage plans; q To be from the distribution One of the prompts in the sampling, namely the power grid status, represents a snapshot of the current power grid status; Represents the old strategy model; This indicates that the old multi-agent model is based on prompts. q A set of G completions is generated, representing G distinct candidate power outage plans; Indicates the first i A sequence of power outage plans; O To complete the sequence of power outage plans; This represents the average value of each completed set. This represents the average value of each set of power outage plans; Indicates the first i The power outage plan sequence is in the first t The action selected at each time step; This indicates that the old strategy model generates actions. The probability of; This indicates that the new policy model generates actions. The probability of; This indicates that the old and new strategies apply to the same action. The proportion of change in preferences; Indicates the relative advantage estimate of the group; The function is used to limit the value between arrive between; The regularization coefficient is used. KL divergence is used to measure the new strategy. Reference Strategy The differences between them; New computational strategies Reference Strategy The KL divergence between them is used as a penalty term, and its expression is: in, Indicates reference strategy Generate Actions The probability of; By continuously adjusting the parameters of the policy model to maximize the GRPO objective function, the training loop terminates when the preset maximum number of iterations is reached, and the final policy model is the trained multi-agent model.

6. The adaptive generation system for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 2, characterized in that, In the strategy generation module, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment based on actual power grid data to generate an optimal time-sharing outage strategy, specifically used for: Based on actual power grid data, and according to the state space defined in the training process. The process involves cleaning, integrating, and standardizing the data to construct state vectors. These state vectors are then input into a trained multi-agent model, which directly outputs the optimal action probability distribution for the current state. Based on this optimal action probability distribution, the power grid system selects the action space with the highest probability. According to the action space The outage time, duration, and maintenance resource allocation of power grid equipment are dynamically adjusted to generate a preliminary outage strategy. This preliminary outage strategy is then quickly verified in a power grid power flow simulation environment with centralized constraints. If the verification passes, the optimal time-sharing outage strategy is obtained.

7. An adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm, characterized in that, Includes the following steps: Based on Markov decision processes, we define the set of agents, state space, action space, reward function, and state transition function, and establish a multi-agent model by combining the objective function. Constraints are constructed and a set of constraints is formed by establishing the power grid operation's safety, supply, and consumption targets respectively. Based on the set of constraints, the multi-agent model is iteratively optimized and trained using a group relative policy optimization algorithm to generate a trained multi-agent model. Based on actual power grid data, the trained multi-agent model dynamically adjusts the outage time, outage duration, and maintenance resource allocation of power grid equipment to generate the optimal time-sharing outage strategy.

8. The adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 7, characterized in that, A multi-agent model is established by combining the objective function, including: The expression for the objective function J is: in, The discount factor is 0 < <1; Let represent the expectation operator, and let represent the optimization of minimizing policy π. As a weighting factor for safety risk scoring, for t Safety score at all times For supply risk scoring weights, for t Supply score at any moment To absorb the risk scoring weights, for t The score for absorption at any given moment; The initial state of the power grid is input, and the decision-making strategies of the agents are initialized. The current state of the power grid is input to each agent. Each agent selects an action from the action space based on its corresponding power grid equipment attributes and the current state of the power grid. The actions of all agents are executed, the maintenance plan matrix is ​​updated, and the new state of the power grid is calculated through the power flow simulation model. Based on the difference between the new state of the power grid and the current state of the power grid, the reward value is calculated through the reward function and fed back to all agents. The agents adjust their decision-making strategies according to the reward value. The above process is iterated until the objective function converges, and the multi-agent model is completed.

9. The adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 7, characterized in that, Constraints are constructed and formed into constraint sets based on the power grid operation's objectives of ensuring safety, supply, and consumption, including: The safety objectives of power grid operation are used to constrain the operational boundaries of power grid equipment, including: Power flow constraints are used to limit the power flow values ​​of a line: ,in, For a moment t line l Trend value, For the line l Permissible power flow limits; Node voltage constraints are used to limit the voltage value at each circuit node in a power grid: ,in, For a moment t line l voltage value, For the circuit nodes of the power grid equipment at time t line l The lower limit of voltage, For the circuit nodes of the power grid equipment at time t line l The upper limit of voltage; The power supply guarantee objective of power grid operation is used to constrain the power supply reliability of power grid equipment, minimizing the number and duration of power outages for users. For power grid equipment whose maintenance would cause repeated power outages to the same area, the corresponding power grid equipment will be scheduled for maintenance within the same time period. ,in, Indicates the first i Planned maintenance time for each piece of equipment. Indicates the first j Planned maintenance time for each piece of equipment; The grid operation's consumption guarantee targets are used to constrain the consumption of new energy sources and the flexibility of the grid system, including: Mutual exclusion constraints are used to stagger the maintenance times of grid equipment that would weaken the grid system or cause additional power outages during maintenance. ; The curtailment constraint is used to ensure that at least one connecting line of a renewable energy unit is in normal operation. If all outgoing lines are scheduled to be out of power, the renewable energy unit will be forced to curtail power. The constraint is defined as the number of non-outage states of the relevant lines being greater than zero. ,in, This represents the total number of lines in a non-power-outage state among all the connection lines of the new energy generating unit; Finally, all the above constraints are combined into a constraint set.

10. The adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm according to claim 7, characterized in that, Based on the constraint set, the multi-agent model is iteratively optimized and trained using a group relative policy optimization algorithm to generate a trained multi-agent model, including: Initialize the parameters of the multi-agent model and embed the complete constraint rules constructed based on the constraint condition set into the power grid power flow simulation environment; The GRPO algorithm is used to perform multiple iterations on the multi-agent model. For each iteration: A batch of power grid states are randomly selected, and G complete power outage plan sequences are generated for each power grid state using a fixed old strategy. Each generated power outage plan sequence is input into the power grid power flow simulation environment. The power grid power flow simulation environment calls the constraint condition set to verify whether each power outage plan violates any constraints, and simulates and deduces the changes in the power grid state after executing the power outage plan. Based on the simulation results and the reward function, the total reward for each power outage plan sequence is calculated. The GRPO objective function is calculated based on the data in the defined tuple, and its expression is: in, This represents the objective function that needs to be maximized; The parameters represent the multi-agent model. It represents the mathematical expectation over all possible grid states and possible outage plans; q To be from the distribution One of the prompts in the sampling, namely the power grid status, represents a snapshot of the current power grid status; Represents the old strategy model; This indicates that the old multi-agent model is based on prompts. q A set of G completions is generated, representing G distinct candidate power outage plans; Indicates the first i A sequence of power outage plans; O To complete the sequence of power outage plans; This represents the average value of each completed set. This represents the average value of each set of power outage plans; Indicates the first i The power outage plan sequence is in the first t The action selected at each time step; This indicates that the old strategy model generates actions. The probability of; This indicates that the new policy model generates actions. The probability of; This indicates that the old and new strategies apply to the same action. The proportion of change in preferences; Indicates the relative advantage estimate of the group; The function is used to limit the value between arrive between; The regularization coefficient is used. KL divergence is used to measure the new strategy. Reference Strategy The differences between them; New computational strategies Reference Strategy The KL divergence between them is used as a penalty term, and its expression is: in, Indicates reference strategy Generate Actions The probability of; By continuously adjusting the parameters of the policy model to maximize the GRPO objective function, the training loop terminates when the preset maximum number of iterations is reached, and the final policy model is the trained multi-agent model.

11. A computer storage medium, wherein the computer-readable storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive generation method for time-sharing power outage strategies of power grid equipment based on the GRPO algorithm as described in any one of claims 7-10.

Images