Distributed Self-Healing Recovery Method and System for Power System Considering Uncertainty of New Energy Output

Abstract

The distributed self-healing restoration method and system for a power system considering the uncertainty of new energy output of the present application relate to the technical field of power systems. By defining the output fluctuation amount and the output volatility rate, setting the triggering conditions for output fluctuation events, and classifying the output fluctuation events; collecting meteorological data and the operation data of new energy units, training a short-term prediction model to predict the new energy output in the future time, judging whether the triggering conditions for output fluctuation events are reached, and if so, evaluating the severity; establishing a self-healing control strategy, designing the state, action, and reward functions of the intelligent agent, and using a reinforcement learning algorithm to optimize the self-healing control strategy; when an output fluctuation event is triggered, the optimized self-healing control strategy is sent to the power system control unit to execute the optimized self-healing control strategy; obtaining the effect evaluation value of the self-healing control of the output fluctuation event, and feeding back the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next round of control strategy.

Description

Technical Field

[0001] This application relates to the technical field of power systems, and particularly to a distributed self-healing recovery method and system for power systems considering the uncertainty of new energy output. Background Art

[0002] Distributed self-healing recovery of power systems means that after a power system fails or is disturbed, it automatically detects and isolates faults in a distributed manner, and quickly restores power supply to make the system return to a normal or near-normal operating state, improving the reliability and recovery ability of the system. This is one of the important contents of building a smart grid.

[0003] A Chinese patent with the publication number CN114465236A discloses a self-healing method for a distribution network to cope with grounding faults and a distribution network. The self-healing method includes fault detection, fault location, fault isolation, and network self-healing; fault detection includes: the protection module of the breaker control device collects the current data of the corresponding power supply line; calculates the current input signal En of the real-time cycle according to the current data; obtains the current change difference ΔFn according to the current input signals En of several cycles; compares the current change differences ΔFn of different power supply lines in the same time period to determine whether the corresponding power supply line has a fault. If so, enter fault location; the distribution network can execute the above self-healing method. The method of this invention is applicable to a global communication intelligent distributed distribution system, and takes corresponding self-healing measures when a small current grounding fault occurs in the distribution system.

[0004] Existing power system self-healing recovery methods do not adequately consider the complexity and uncertainty in the scenario of new energy access, lacking flexibility and self-adaptability. Summary of the Invention

[0005] This application aims to solve at least one of the technical problems in the related art to some extent. For this reason, an object of this application is to propose a distributed self-healing recovery method and system for power systems considering the uncertainty of new energy output, improving the self-adaptability of distributed self-healing recovery of power systems.

[0006] One aspect of this application provides a distributed self-healing recovery method for power systems considering the uncertainty of new energy output, including:

[0007] Step S100: Define the new energy output fluctuation amount and the new energy output volatility rate, set the trigger condition for the new energy output fluctuation event, and classify the new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction;

[0008] Step S200: Collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output at future times, determine whether the trigger condition for the new energy output fluctuation event is reached, and if so, evaluate the severity of the new energy output fluctuation event;

[0009] Step S300: Establish a self-healing control strategy for new energy output fluctuation events, design the state, action, and reward functions of the intelligent agent, and optimize the self-healing control strategy using a reinforcement learning algorithm;

[0010] Step S400: Each control unit of the power system monitors new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy;

[0011] Step S500: Obtain the effect evaluation value of the self-healing control of new energy output fluctuation events, feedback the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategies;

[0012] The specific method for defining the new energy output fluctuation amount and the new energy output volatility rate, setting the trigger condition for the new energy output fluctuation event, and classifying the new energy output fluctuation event according to the fluctuation amplitude, duration, and fluctuation direction is as follows:

[0013] Step S110: Preset a time window mt, define the change amount of new energy output within the time window as the new energy output fluctuation amount Δp, and define the ratio of the new energy output fluctuation amount to the new energy installed capacity as the new energy output volatility rate Ra;

[0014] Step S120: Set the first volatility threshold R th1 . When the new energy output volatility rate is greater than or equal to the first volatility threshold, it is considered that the new energy output fluctuation event is triggered;

[0015] Step S130: According to the magnitude of the new energy output volatility rate, divide the fluctuation amplitude into three levels: small fluctuation, medium fluctuation, and large fluctuation. The range of the new energy output volatility rate for small fluctuations is [R th1 , R th2 ), the range of the new energy output volatility rate for medium fluctuations is [R th2 , R th3 ), and the range of the new energy output volatility rate for large fluctuations is [R th3 , +∞), where R th1 , R th2 , and R th3 are the preset first volatility threshold, second volatility threshold, and third volatility threshold respectively;

[0016] Step S140: According to the duration T of the new energy output fluctuation event dur divide it into short-term fluctuations and long-term fluctuations. The duration range of the short-term fluctuations is (0, T th ), and the duration range of the long-term fluctuations is [T th , +∞), where T th is a preset time threshold;

[0017] Step S150: According to the positive or negative of the new energy output fluctuation amount Δp, divide the fluctuation direction into upward fluctuations and downward fluctuations;

[0018] The specific method of collecting meteorological data and the operation data of new energy units, training a short-term prediction model for predicting new energy output, and using the short-term prediction model to predict the new energy output in the future time and determining whether the trigger condition of the new energy output fluctuation event is reached is as follows:

[0019] Step S210: Configure meteorological sensors and electrical sensors at the new energy station to collect meteorological data and the operation data of new energy units. The meteorological data includes historical meteorological data and weather forecast data, and the operation data includes the inverter output power and the total station output;

[0020] Step S220: Select the long short-term memory network as the initial network of the short-term prediction model, preset the sliding window length nt and the sliding step, and generate a time series from the historical meteorological data, operation data, and weather forecast data in chronological order;

[0021] Step S230: Use the historical meteorological data and operation data at nt moments before the t-th moment and the weather forecast data at nt moments after the t-th moment as the input data of the short-term prediction model, use the future new energy output as the output data, and move the sliding window backward on the time series in turn according to the sliding step to generate training samples in the form of a sliding window on the time series;

[0022] Step S240: Define the mean square error as the loss function for training the long short-term memory network, use the training samples to train the long short-term memory network, optimize the model parameters by minimizing the value of the loss function, and when the loss function converges, obtain the trained short-term prediction model;

[0023] Step S250: Every time window mt, input the historical meteorological data and operation data at nt moments before the current moment and the weather forecast data at nt moments after the current moment into the short-term prediction model, predict the future new energy output, and calculate the new energy output fluctuation amount Δp and the new energy output volatility Ra according to the predicted new energy output;

[0024] Step S260: Compare the new - energy output volatility with the first volatility threshold R th1 If the new - energy output volatility is greater than or equal to the first volatility threshold, trigger a new - energy output fluctuation event;

[0025] The specific method for evaluating the severity of the new - energy output fluctuation event is as follows:

[0026] Step S270: Judge the severity of the fluctuation amplitude, duration, and fluctuation direction of the new - energy output fluctuation event;

[0027] The method for judging the fluctuation amplitude is: Define the severity of the fluctuation amplitude as 1, 2, and 3 for small fluctuations, medium fluctuations, and large fluctuations respectively. According to the magnitude of the new - energy output volatility, judge the value of the severity of the fluctuation amplitude Rs, where Rs = {1, 2, 3};

[0028] The method for judging the duration is: Obtain the new - energy output sequence based on n - times of predicted new - energy output. Judge the moment t1 when the new - energy output volatility Ra is first greater than or equal to the first volatility threshold R th1 as the starting moment of the new - energy output fluctuation event; Starting from the moment t1, judge the moment t2 when the new - energy output volatility Ra is first less than the first volatility threshold R th1 as the ending moment of the new - energy output fluctuation event. Obtain the duration according to the difference between the moment t1 and the moment t2. Define the severity of the duration as 1 and 2 for short - term fluctuations and long - term fluctuations respectively. According to the magnitude of the duration, judge the value of the severity of the duration RT, where RT = {1, 2};

[0029] The method for judging the fluctuation direction is: Define the severity of the fluctuation direction as 1 and 2 for upward fluctuations and downward fluctuations respectively. According to the positive or negative of the new - energy output fluctuation amount, judge the value of the severity of the fluctuation direction Rd, where Rd = {1, 2};

[0030] Step S280: Calculate the severity S of the new - energy output fluctuation event according to the severities of the fluctuation amplitude, duration, and fluctuation direction;

[0031] Step S290: Set severity thresholds S1 and S2, and classify the severity of the new - energy output fluctuation event into mild fluctuations, moderate fluctuations, and severe fluctuations; Among them, the range of the severity of mild fluctuations is: 1 ≤ S < S1, the range of the severity of moderate fluctuations is: S1 ≤ S < S2, and the range of the severity of severe fluctuations is: S2 ≤ S < 3;

[0032] The specific method for designing the state, action, and reward function of the intelligent agent and optimizing the self - healing control strategy using the reinforcement learning algorithm is as follows:

[0033] Step S310: Define the state space of the agent. The state s includes the volatility Ra of new energy output, the grid frequency deviation Δf, the state of charge SOC of the energy storage system, the state of controllable load, and the severity S of the new energy output fluctuation event. Define the action space of the agent. The action a includes the new energy output limit command P cur , the charge and discharge power command P ess of the energy storage, and the controllable load control command P load ;

[0034] Step S320: Define the reward function of the agent based on the grid frequency deviation Δf, the new energy output limit |P cur |, the change in the state of charge of the energy storage system |ΔSOC|, the controllable load control amount |P load |, and the evaluation value of the effect of historical actions;

[0035] Step S330: Adopt a reinforcement learning algorithm to initialize the current Q-network parameters. At the state s t at the current moment, randomly select an action with probability ε, or select the action with the largest Q value with probability 1 - ε, and execute the selected action a t , to obtain the next moment state s t+1 and the reward function value r t ;

[0036] Step S340: Store s t , a t , r t , s t+1 in the experience replay buffer, randomly extract a batch of samples s j , a j , r j , s j+1 from the experience replay buffer, and calculate the target value of the current Q-network;

[0037] Step S350: Define the loss function L(θ) of the current Q-network, update the current Q-network parameters θ by gradient descent, minimize the value of the loss function, and periodically update the target Q-network parameters θ - to the current Q-network parameters θ;

[0038] Step S360: The agent selects the optimal action according to the current state. When selecting an action, if an action outside the constraint range is selected, reselect until an action within the constraint range is selected, to obtain an optimized self-healing control strategy;

[0039] The specific method for establishing the self-healing control strategy for new energy output fluctuation events is as follows:

[0040] Set the self-healing control strategy according to the severity of the new energy output fluctuation event, which is the new energy output limit command P in the action space cur , the energy storage charge and discharge power command P ess and the controllable load control command P load constraint range; among them, when the fluctuation is mild, the constraint range of the new energy output limit command is P cur ∈[0.9P0, P0], the constraint range of the energy storage charge and discharge power command is the constraint range of the controllable load control command is P load ∈[0.95P l , 1.05P l ; when the fluctuation is moderate, the constraint range of the new energy output limit command is P cur ∈[0.8P0, 0.9P0), the constraint range of the energy storage charge and discharge power command is the constraint range of the controllable load control command is P load ∈[0.9P l , 1.1P l ; when the fluctuation is severe, the constraint range of the new energy output limit command is P cur ∈[0.7P0, 0.8P0), the constraint range of the energy storage charge and discharge power command is the constraint range of the controllable load control command is P load ∈[0.85P l , 1.15P l , where P0 is the initial new energy output, is the rated power of the energy storage system, and P l is the initial power of the controllable load;

[0041] Each control unit of the power system monitors the new energy output fluctuation event in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system. The specific method for executing the optimized self-healing control strategy is as follows:

[0042] Step S410: Install monitoring equipment and communication equipment on each control unit of the power system to monitor the new energy output fluctuation event in real time. When a control unit detects a new energy output fluctuation event in its jurisdiction, the optimized self-healing control strategy is sent to each control unit of the power system, and according to the new energy output limit command, the energy storage charge and discharge power command, and the controllable load control command, each control unit is adjusted to execute the corresponding command;

[0043] The specific method for obtaining the effect evaluation value of the self-healing control of the new energy output fluctuation event and feeding the effect evaluation value back to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy is as follows:

[0044] Step S510: After executing the optimized self-healing control strategy, obtain the effect evaluation index of the self-healing control. The effect evaluation index includes: the grid frequency recovery time T fr , the energy storage cycle life loss L ess and the controllable load regulation cost C load ;

[0045] Step S520: Calculate the effect evaluation value of the self-healing control of the new energy output fluctuation event according to the effect evaluation index;

[0046] Step S530: Feed back the effect evaluation value obtained by the current execution of the self-healing control strategy to the reinforcement learning algorithm for guiding the optimization of the next self-healing control strategy.

[0047] An aspect of the present application provides a distributed self-healing recovery system for a power system considering the uncertainty of new energy output, including:

[0048] A fluctuation event classification module, used to define the new energy output fluctuation amount and the new energy output volatility rate, set the trigger condition of the new energy output fluctuation event, and classify the new energy output fluctuation event according to the fluctuation amplitude, duration and fluctuation direction;

[0049] A severity calculation module, used to collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output in the future, judge whether the trigger condition of the new energy output fluctuation event is reached, and if so, evaluate the severity of the new energy output fluctuation event;

[0050] A self-healing strategy optimization module, used to establish a self-healing control strategy for the new energy output fluctuation event, design the state, action and reward functions of the intelligent agent, and optimize the self-healing control strategy by using the reinforcement learning algorithm;

[0051] A self-healing strategy execution module, used for each control unit of the power system to monitor the new energy output fluctuation event in real time. When the new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy;

[0052] An effect evaluation feedback module, used to obtain the effect evaluation value of the self-healing control of the new energy output fluctuation event, and feed back the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy.

[0053] One aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for distributed self-healing recovery of a power system considering the uncertainty of new energy output.

[0054] One aspect of the present application provides a readable storage medium storing a computer program, which is suitable for being loaded by a processor to execute the steps in the method for distributed self-healing recovery of a power system considering the uncertainty of new energy output.

[0055] The method and system for distributed self-healing recovery of a power system considering the uncertainty of new energy output proposed in the present application have the following advantages compared with the prior art:

[0056] The present application clearly defines indicators such as the fluctuation amount and volatility of new energy output, and comprehensively defines and quantifies new energy output fluctuation events from multiple dimensions including the fluctuation amplitude, duration, and direction.

[0057] The present application uses deep learning methods such as LSTM, combines meteorological data and operation data to train a short-term prediction model, and can predict the new energy output fluctuation trend in advance to gain time for taking countermeasures in advance.

[0058] The present application comprehensively considers factors such as the fluctuation amplitude, duration, and direction, and constructs a quantitative index for the severity of fluctuation events. This helps to formulate differentiated self-healing control strategies according to the severity level of fluctuations and improve the pertinence of measures.

[0059] The present application models the self-healing control problem as a Markov decision process, designs the agent state, action, and reward functions, and uses reinforcement learning algorithms to solve the optimal control strategy. The control process can adapt to different severity fluctuation scenarios, reflecting the autonomous decision-making ability based on artificial intelligence.

[0060] The present application feeds back the actual control effect evaluation result to the reinforcement learning algorithm to guide the optimization of the next round of control strategy. Through the closed loop of "prediction - evaluation - optimal control - effect feedback - re-optimization", the effect of self-healing control can be continuously improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is the flowchart of the method for distributed self-healing recovery of a power system considering the uncertainty of new energy output provided by the present application;

[0062] Figure 2 is the flowchart of the method for classifying new energy output fluctuation events provided by the present application;

[0063] Figure 3Flowchart of the new energy output fluctuation event triggering method provided by this application;

[0064] Figure 4 Functional module diagram of the power system distributed self-healing recovery system considering the uncertainty of new energy output provided by this application. Detailed implementation manners

[0065] To better understand this application, various aspects of this application will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of the exemplary embodiments of this application and do not limit the scope of this application in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0066] In the drawings, for ease of illustration, the size, dimensions, and shape of the elements have been slightly adjusted. The drawings are only examples and are not drawn to an exact scale. As used herein, terms such as "substantially", "about", and similar terms are used as terms of approximation and not as terms of degree, and are intended to account for the inherent deviations in measured or calculated values that would be recognized by a person of ordinary skill in the art. Additionally, in this application, the order in which the steps are described does not necessarily represent the order in which these processes occur in actual operation, unless otherwise clearly specified or derivable from the context.

[0067] It should also be understood that expressions such as "comprises", "comprising", "has", "including", and / or "including having" are open-ended rather than closed-ended expressions in this specification, which mean that there are the stated features, elements, and / or components, but do not exclude the existence of one or more other features, elements, components, and / or their combinations. In addition, when an expression such as "at least one of..." appears after a list of listed features, it modifies the entire list of features rather than just an individual element in the list. Further, when describing the embodiments of this application, the use of "may" means "one or more embodiments of this application". And the term "exemplary" is intended to refer to an example or illustration.

[0068] Unless otherwise defined, all terms used herein (including engineering terms and scientific and technical terms) have the same meaning as commonly understood by a person of ordinary skill in the art to which this application belongs. It should also be understood that, unless clearly stated in this application, words defined in a common dictionary should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and should not be interpreted in an idealized or overly formal sense.

[0069] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0070] Embodiment 1

[0071] As Figure 1 shown, it is a distributed self-healing recovery method for a power system considering the uncertainty of new energy output provided by the present application, including:

[0072] Step S100: Define the new energy output fluctuation amount and the new energy output volatility rate, set the trigger condition for the new energy output fluctuation event, and classify the new energy output fluctuation event according to the fluctuation amplitude, duration, and fluctuation direction;

[0073] The specific method for defining the new energy output fluctuation amount and the new energy output volatility rate, setting the trigger condition for the new energy output fluctuation event, and classifying the new energy output fluctuation event according to the fluctuation amplitude, duration, and fluctuation direction is as follows:

[0074] Step S110: Preset a time window mt, define the change amount of the new energy output within the time window as the new energy output fluctuation amount Δp, and define the ratio of the new energy output fluctuation amount to the new energy installed capacity as the new energy output volatility rate Ra;

[0075] The new energy output fluctuation amount represents the change amount of the new energy output within the time window mt;

[0076] The new energy output volatility rate represents the ratio of the new energy output fluctuation amount Δp to the new energy installed capacity p cap That is,

[0077] Step S120: Set a first volatility threshold R th1 , when the new energy output volatility rate is greater than or equal to the first volatility threshold, it is considered that the new energy output fluctuation event is triggered;

[0078] Step S130: According to the magnitude of the new energy output volatility rate, divide the fluctuation amplitude into three levels: small fluctuation, medium fluctuation, and large fluctuation. The range of the new energy output volatility rate for small fluctuation is [R th1 , R th2 ), the range of the new energy output volatility rate for medium fluctuation is [R th2 , R th3 ), and the range of the new energy output volatility rate for large fluctuation is [R th3 , +∞), where R th1 , R th2 , R th3They are respectively a preset first volatility threshold, a second volatility threshold, and a third volatility threshold;

[0079] Step S140: According to the duration T of the new energy output fluctuation event dur Classify it into short-term fluctuations and long-term fluctuations. The duration range of the short-term fluctuations is (0, T th ), and the duration range of the long-term fluctuations is [T th , +∞), where T th is a preset time threshold;

[0080] Step S150: According to the positive or negative of the new energy output fluctuation amount Δp, divide the fluctuation direction into upward fluctuations and downward fluctuations;

[0081] Figure 2 is the flow chart of the new energy output fluctuation event classification method provided by this application;

[0082] Step S200: Collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output at future times, and determine whether the trigger condition of the new energy output fluctuation event is reached. If it is reached, evaluate the severity of the new energy output fluctuation event;

[0083] The specific method of collecting meteorological data and the operation data of new energy units, training a short-term prediction model for predicting new energy output, using the short-term prediction model to predict the new energy output at future times, and determining whether the trigger condition of the new energy output fluctuation event is reached is as follows:

[0084] Step S210: Configure meteorological sensors and electrical sensors at the new energy station to collect meteorological data and the operation data of new energy units. The meteorological data includes historical meteorological data and weather forecast data, and the operation data includes the inverter output power and the total output of the station;

[0085] The historical meteorological data includes the wind speed, radiation intensity, and temperature at historical times;

[0086] The weather forecast data includes the wind speed, radiation intensity, and temperature at future times predicted by the meteorological station;

[0087] Step S220: Select a long short-term memory network as the initial network of the short-term prediction model, preset the sliding window length nt and the sliding step, and generate a time series of historical meteorological data, operation data, and weather forecast data in chronological order;

[0088] The values of the sliding window length nt and the sliding step are preset by those skilled in the art according to experience;

[0089] Step S230: Use the historical meteorological data and operation data of nt moments before the t-th moment, the weather forecast data of nt moments after the t-th moment as the input data of the short-term prediction model, use the future new energy output as the output data, and the sliding window moves backward in the time series in turn according to the sliding step length, and generates training samples in the time series in the way of the sliding window;

[0090] Step S240: Define the mean square error as the loss function for training the long short-term memory network, use the training samples to train the long short-term memory network, optimize the model parameters by minimizing the value of the loss function, and when the loss function converges, obtain the trained short-term prediction model;

[0091] The calculation formula of the loss function is: where, N yb is the total number of training samples, P n is the actual new energy output in the n-th training sample, is the predicted new energy output;

[0092] Step S250: Every time window mt, input the historical meteorological data and operation data of nt moments before the current moment, and the weather forecast data of nt moments after that into the short-term prediction model, predict the future new energy output, and calculate the new energy output fluctuation amount Δp and the new energy output volatility Ra according to the predicted new energy output;

[0093] The calculation formula of the new energy output fluctuation amount is: Δp = P t -P t-mt where, P t represents the new energy output at the current moment t, P t-mt represents the new energy output at the moment t-mt one time window before, and Δp is the new energy output fluctuation amount between the two moments;

[0094] Step S260: Compare the new energy output volatility with the first volatility threshold R th1 If the new energy output volatility is greater than or equal to the first volatility threshold, trigger the new energy output fluctuation event;

[0095] The duration is judged according to the predicted new energy output;

[0096] Figure 3 This is the flow chart of the new energy output fluctuation event triggering method provided by this application;

[0097] The specific method for evaluating the severity of the new energy output fluctuation event is:

[0098] Step S270: Judge the severity of the fluctuation amplitude, duration and fluctuation direction of the new energy output fluctuation event;

[0099] The method for judging the fluctuation amplitude is as follows: Define the severity of the fluctuation amplitude as 1, 2, and 3 for small fluctuations, medium fluctuations, and large fluctuations respectively. According to the magnitude of the new energy output volatility, judge the value of the severity Rs of the fluctuation amplitude, where Rs = {1, 2, 3};

[0100] The method for judging the duration is as follows: Obtain the new energy output sequence based on the new energy output predicted n times, and judge the moment t1 when the new energy output volatility Ra is greater than or equal to the first volatility threshold R for the first time as the starting moment of the new energy output fluctuation event; Starting from the moment t1, judge the moment t2 when the new energy output volatility Ra is less than the first volatility threshold R for the first time thereafter as the ending moment of the new energy output fluctuation event, and obtain the duration according to the difference between the moment t1 and the moment t2; Define the severity of the duration as 1 and 2 for short-term fluctuations and long-term fluctuations respectively. According to the magnitude of the duration, judge the value of the severity RT of the duration, where RT = {1, 2}; th1 of as the starting moment of the new energy output fluctuation event; Starting from the moment t1, judge the moment t2 when the new energy output volatility Ra is less than the first volatility threshold R for the first time thereafter as the ending moment of the new energy output fluctuation event, and obtain the duration according to the difference between the moment t1 and the moment t2; Define the severity of the duration as 1 and 2 for short-term fluctuations and long-term fluctuations respectively. According to the magnitude of the duration, judge the value of the severity RT of the duration, where RT = {1, 2}; th1 The calculation formula for the duration is: T

[0101] The calculation formula for the duration is: T dur = t2 - t1;

[0102] The method for judging the fluctuation direction is as follows: Define the severity of the fluctuation direction as 1 and 2 for upward fluctuations and downward fluctuations respectively. According to the positive and negative of the new energy output fluctuation amount, judge the value of the severity Rd of the fluctuation direction, where Rd = {1, 2};

[0103] Step S280: Calculate the severity S of the new energy output fluctuation event according to the severities of the fluctuation amplitude, duration, and fluctuation direction;

[0104] The calculation formula for the severity of the new energy output fluctuation event is: S = ws × Rs + wT × RT + wd × Rd, where ws, wT, and wd are the weight coefficients of the severities of the fluctuation amplitude, duration, and fluctuation direction respectively, and are set by those skilled in the art according to experience;

[0105] Step S290: Set the severity thresholds S1 and S2, and classify the severities of the new energy output fluctuation events into mild fluctuations, moderate fluctuations, and severe fluctuations; Among them, the range of the severity of mild fluctuations is: 1 ≤ S < S1, the range of the severity of moderate fluctuations is: S1 ≤ S < S2, and the range of the severity of severe fluctuations is: S2 ≤ S < 3;

[0106] Step S300: Establish a self-healing control strategy for the new energy output fluctuation event, design the state, action, and reward functions of the intelligent agent, and optimize the self-healing control strategy by using a reinforcement learning algorithm;

[0107] The specific method for establishing a self-healing control strategy for new energy output fluctuation events, designing the state, action, and reward functions of the agent, and optimizing the self-healing control strategy using a reinforcement learning algorithm is as follows:

[0108] Step S310: Define the state space of the agent. The state s includes the new energy output volatility Ra, the grid frequency deviation Δf, the state of charge SOC of the energy storage system, the controllable load state, and the severity S of the new energy output fluctuation event. Define the action space of the agent. The action a includes the new energy output limit command P cur , the energy storage charge and discharge power command P ess and the controllable load control command P load ;

[0109] The grid frequency deviation reflects the degree to which the grid frequency deviates from the rated value and is an indicator for measuring the grid stability. The grid frequency is measured in real time by wide-area measurement devices deployed at key grid nodes, and its deviation degree is calculated;

[0110] The state of charge of the energy storage system represents the ratio of the current available capacity of the energy storage system to the rated capacity, which affects the regulation ability of the energy storage system and is obtained through real-time monitoring by the energy storage battery management system;

[0111] The controllable load state refers to the access power level of the controllable load, which affects the regulation ability of the load-side resources. The real-time power of the controllable load is monitored by the load management system, and the percentage of the real-time power of the controllable load to the rated power of the controllable load is calculated according to the rated power of the controllable load;

[0112] The new energy output limit command means that when the new energy output is unstable, the output limit of the new energy power station is adjusted to reduce the new energy output fluctuation amplitude and relieve the grid frequency fluctuation;

[0113] The energy storage charge and discharge power command means to adjust the charge and discharge power of the energy storage system, smooth the new energy output fluctuation, and maintain the grid frequency stability;

[0114] The controllable load control command means to adjust the access power of the controllable load, smooth the new energy output fluctuation, and maintain the grid frequency stability;

[0115] Step S320: Define the reward function of the agent based on the grid frequency deviation Δf, the new energy output limit |P cur |, the change in the state of charge of the energy storage system |ΔSOC|, the controllable load control quantity |P load |, and the effect evaluation value of the historical action;

[0116] The calculation formula of the reward function is: r = -ω1×|Δf| - ω2×|P cur|-ω3×|ΔSOC|-ω4×|P load |+ω5×R zb , where R zb is the effect evaluation value, and ω1, ω2, ω3, ω4, and ω5 are the weight coefficients of the grid frequency deviation, new energy output limit, charge change of the energy storage system, controllable load control amount, and effect evaluation value of historical actions respectively;

[0117] The weight coefficients of the grid frequency deviation, new energy output limit, charge change of the energy storage system, controllable load control amount, and effect evaluation value of historical actions are set by those skilled in the art according to experience;

[0118] The new energy output limit refers to the difference in new energy output before and after executing the current action during the fluctuation of new energy output;

[0119] The charge change of the energy storage system refers to the absolute value of the change in the state of charge SOC of the energy storage system before and after executing the current action during the fluctuation of new energy output;

[0120] The controllable load control amount represents the absolute value of the control amplitude of the controllable load control command before and after executing the current action during the fluctuation of new energy output;

[0121] Step S330: Adopt a reinforcement learning algorithm to initialize the parameters of the current Q-network. At the state s t at the current moment, randomly select an action with probability ε, or select the action with the largest Q value with probability 1 - ε, and execute the selected action a t , to obtain the state s t+1 at the next moment and the reward function value r t ;

[0122] Preferably, the reinforcement learning algorithm selects the DQN algorithm;

[0123] The value of probability ε is set by those skilled in the art according to experience;

[0124] The DQN algorithm includes two Q-networks with the same structure but different parameters: the current Q-network and the target Q-network. The current Q-network is used for action selection and parameter update, and the target Q-network is used for calculating the target value, and its parameters θ - are copied from the current Q-network every certain number of steps.

[0125] Step S340: Store s t , a t , r t , s t+1 into the experience replay buffer, randomly extract a batch of samples s j , a j , rj , s j+1 , calculate the target value of the current Q-network;

[0126] The calculation formula for the target value of the current Q-network is: where γ is the discount factor, θ - is the parameter of the target Q-network, a’ represents the action selected under state s j+1 , Q(s j+1 , a’; θ - ) represents the action value function calculated by the target Q-network, used to estimate the Q-value when taking action a’ under state s j+1 , max a’ represents selecting the action that can obtain the maximum value among all actions a’;

[0127] Step S350: Define the loss function L(θ) of the current Q-network, update the parameters θ of the current Q-network by gradient descent method, minimize the value of the loss function, and periodically update the parameters θ - of the target Q-network to the parameters θ of the current Q-network;

[0128] The calculation formula for the loss function of the current Q-network is: where represents taking the expected value, Q(s j , a j ; θ) represents the predicted value of the current Q-network, and the loss function L(θ) of the current Q-network represents the mean square error between the predicted value and the target value of the current Q-network;

[0129] Step S360: Set the self-healing control strategy according to the severity of the new energy output fluctuation event, as the constraint range of the new energy output limit instruction P cur , the energy storage charge and discharge power instruction P ess and the controllable load control instruction P load in the action space; among them, when the fluctuation is mild, the constraint range of the new energy output limit instruction is P cur ∈ [0.9P0, P0], the constraint range of the energy storage charge and discharge power instruction is the constraint range of the controllable load control instruction is P load ∈ [0.95P l , 1.05P l ; when the fluctuation is moderate, the constraint range of the new energy output limit instruction is P cur ∈ [0.8P0, 0.9P0), the constraint range of the energy storage charge and discharge power instruction is the constraint range of the controllable load control instruction is P load ∈ [0.9Pl , 1.1P l ; When there is severe fluctuation, the constraint range of the new - energy output limit instruction is P cur ∈[0.7P0, 0.8P0), the constraint range of the energy - storage charge - discharge power instruction is the constraint range of the controllable - load control instruction is P load ∈[0.85P l , 1.15P l , where P0 is the initial output of new energy, is the rated power of the energy - storage system, P l is the initial power of the controllable load;

[0130] Step S370: The agent selects the optimal action according to the current state. When selecting an action, if an action outside the constraint range is selected, re - select until an action within the constraint range is selected, and an optimized self - healing control strategy is obtained;

[0131] Step S400: Each control unit of the power system monitors new - energy output fluctuation events in real - time. When a new - energy output fluctuation event is triggered, the optimized self - healing control strategy is sent to the distributed control unit of the power system, and the optimized self - healing control strategy is executed;

[0132] The specific method for each control unit of the power system to monitor new - energy output fluctuation events in real - time, and when a new - energy output fluctuation event is triggered, to send the optimized self - healing control strategy to the distributed control unit of the power system and execute the optimized self - healing control strategy is as follows:

[0133] Step S410: Monitoring devices and communication devices are installed in each control unit of the power system to monitor new - energy output fluctuation events in real - time. When a control unit monitors that a new - energy output fluctuation event is triggered within its jurisdiction, the optimized self - healing control strategy is sent to each control unit of the power system, and according to the new - energy output limit instruction, the energy - storage charge - discharge power instruction, and the controllable - load control instruction, each control unit is adjusted to execute the corresponding instruction;

[0134] Step S500: Obtain the effect evaluation value of the self - healing control of the new - energy output fluctuation event, and feedback the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next - round self - healing control strategy;

[0135] The specific method for obtaining the effect evaluation value of the self - healing control of the new - energy output fluctuation event, and feedbacking the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next - round self - healing control strategy is as follows:

[0136] Step S510: After executing the optimized self - healing control strategy, obtain the effect evaluation index of the self - healing control. The effect evaluation index includes: the grid - frequency recovery time T fr, energy storage cycle life loss L ess and controllable load regulation cost C load ;

[0137] The power grid frequency recovery time refers to the time used for the power grid frequency deviation Δf to return to the normal range after the fluctuation occurs. The frequency data is recorded in real time by the power grid frequency monitoring device, and the recovery time is judged according to the normal range; the normal range is determined according to the power grid frequency standard and the actual operation requirements. Preferably, it can be set to ±0.05Hz of the rated frequency;

[0138] The energy storage cycle life loss refers to the ratio of the number of energy storage charge and discharge cycles to the rated cycle life. The number of charge and discharge cycles is recorded according to the change of the SOC of the energy storage system, and the energy storage cycle life loss is calculated in combination with the rated cycle life. The rated cycle life is determined according to the type of energy storage battery and the data provided by the manufacturer;

[0139] The controllable load regulation cost refers to the user utility loss and compensation cost caused by the controllable load regulation. The regulation cost is calculated according to the type and regulation power of the controllable load, in combination with the utility function and the compensation cost;

[0140] The calculation formula for the controllable load regulation cost is: where K is the total number of controllable loads, U k (·) is the utility function of the kth controllable load, are the controllable load powers before and after regulation respectively, is the compensation cost of the kth controllable load;

[0141] The function expression of the utility function of the controllable load is: where U k (P k ) represents the user utility value of the kth controllable load at the controllable load power P k . α and β are the parameters of the utility function, representing the user's preference for electric energy, which are set by those skilled in the art according to experience, α>0, β>0;

[0142] Step S520: Calculate the effect evaluation value of the self-healing control of the new energy output fluctuation event according to the effect evaluation index;

[0143] The calculation formula for the effect evaluation value of the self-healing control of the new energy output fluctuation event is: R zb =-ω fr ×T fr -ω ess ×L ess -ω load ×C load , where ωfr and ω ess and ω load are the weight coefficients of the grid frequency recovery time, the energy storage cycle life loss, and the controllable load regulation cost respectively;

[0144] The weight coefficients of the grid frequency recovery time, the energy storage cycle life loss, and the controllable load regulation cost are set by those skilled in the art according to actual needs, and the sum of the weight coefficients is 1;

[0145] Step S530: Feed back the effect evaluation value obtained by executing the current self-healing control strategy to the reinforcement learning algorithm for guiding the optimization of the next self-healing control strategy.

[0146] Embodiment 2

[0147] As Figure 4 shown, the distributed self-healing recovery system for a power system considering the uncertainty of new energy output provided by this application includes:

[0148] A fluctuation event classification module, which is used to define the new energy output fluctuation amount and the new energy output volatility rate, set the trigger conditions for new energy output fluctuation events, and classify the new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction;

[0149] A severity calculation module, which is used to collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output in the future, judge whether the trigger conditions for new energy output fluctuation events are reached, and if so, evaluate the severity of the new energy output fluctuation events;

[0150] A self-healing strategy optimization module, which is used to establish a self-healing control strategy for new energy output fluctuation events, design the state, action, and reward functions of the intelligent agent, and optimize the self-healing control strategy by using a reinforcement learning algorithm;

[0151] A self-healing strategy execution module, which is used for each control unit of the power system to monitor new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy;

[0152] An effect evaluation feedback module, which is used to obtain the effect evaluation value of the self-healing control of new energy output fluctuation events, and feed back the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy.

[0153] Embodiment 3

[0154] According to another aspect of the present application, an electronic device is also provided. The electronic device may include one or more processors and one or more memories. Among them, computer-readable code is stored in the memory, and when the computer-readable code is run by one or more processors, it can execute the distributed self-healing recovery method for a power system considering the uncertainty of new energy output as described above.

[0155] The method or system according to the embodiments of the present application can also be implemented by means of the following architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, input / output components, a hard disk, etc. The storage device in the electronic device, such as ROM or hard disk, can store the distributed self-healing recovery method for a power system considering the uncertainty of new energy output provided by the present application. The distributed self-healing recovery method for a power system considering the uncertainty of new energy output may, for example, include: defining the new energy output fluctuation amount and the new energy output volatility rate, setting the trigger conditions for new energy output fluctuation events, classifying new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction; collecting meteorological data and the operation data of new energy units, training a short-term prediction model for predicting new energy output, using the short-term prediction model to predict the new energy output at future times, and determining whether the trigger conditions for new energy output fluctuation events are met. If so, evaluating the severity of the new energy output fluctuation events; establishing a self-healing control strategy for new energy output fluctuation events, designing the state, action, and reward functions of the intelligent agent, and optimizing the self-healing control strategy using a reinforcement learning algorithm; each control unit of the power system monitors new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system, and the optimized self-healing control strategy is executed; obtaining the effect evaluation value of the self-healing control of the new energy output fluctuation event, and feeding the effect evaluation value back to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy. Further, the electronic device may also include a user interface. Of course, the above architecture is only exemplary. When implementing different devices, one or more components of the above electronic device can be omitted according to actual needs.

[0156] Embodiment 4

[0157] A readable storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are run by a processor, a distributed self-healing recovery method for a power system considering the uncertainty of new energy output as described with reference to the above figures can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disks, flash memory, etc.

[0158] In addition, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present application, such as: defining the fluctuation amount and volatility of new energy output, setting the trigger conditions for new energy output fluctuation events, classifying new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction; collecting meteorological data and the operating data of new energy units, training a short-term prediction model for predicting new energy output, using the short-term prediction model to predict the new energy output at future times, and determining whether the trigger conditions for new energy output fluctuation events are met. If so, evaluating the severity of the new energy output fluctuation events; establishing a self-healing control strategy for new energy output fluctuation events, designing the states, actions, and reward functions of the agent, and optimizing the self-healing control strategy using a reinforcement learning algorithm; each control unit of the power system monitors new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy; obtaining the effect evaluation value of the self-healing control of the new energy output fluctuation event and feeding the effect evaluation value back to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0159] The methods, devices, and equipment of the present application can be implemented in many ways. For example, the methods, devices, and equipment of the present application can be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is only for illustration, and the steps of the method of the present application are not limited to the specific order described above, unless otherwise specifically stated. In addition, in some embodiments, the present application can also be implemented as a program recorded on a recording medium, and these programs include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers a recording medium storing a program for executing the method according to the present application.

[0160] In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0161] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A distributed self-healing recovery method for a power system considering the uncertainty of new energy output, characterized in that, Including: Define the new energy output fluctuation amount and the new energy output volatility rate, set the trigger conditions for new energy output fluctuation events, and classify new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction; Collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output at future times, and determine whether the trigger conditions for new energy output fluctuation events are met. If so, evaluate the severity of the new energy output fluctuation events; Establish a self-healing control strategy for new energy output fluctuation events, design the states, actions, and reward functions of the intelligent agent, and use the reinforcement learning algorithm to optimize the self-healing control strategy; Each control unit of the power system monitors new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy; Obtain the effect evaluation value of the self-healing control of new energy output fluctuation events, and feedback the effect evaluation value to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy.

2. The distributed self-healing recovery method for a power system considering the uncertainty of new energy output according to claim 1, characterized in that The specific method for defining the new energy output fluctuation amount and the new energy output volatility rate, setting the trigger conditions for new energy output fluctuation events, and classifying new energy output fluctuation events according to the fluctuation amplitude, duration, and fluctuation direction is as follows: Preset a time window \(m_t\), define the change amount of new energy output within the time window as the new energy output fluctuation amount \(\Delta p\), and define the ratio of the new energy output fluctuation amount to the new energy installed capacity as the new energy output volatility rate \(R_a\); Set the first volatility threshold R th1 , when the volatility of new energy output is greater than or equal to the first volatility threshold, it is considered that a new energy output fluctuation event has been triggered; According to the magnitude of the new - energy output volatility, the fluctuation range is divided into three levels: small - fluctuation, medium - fluctuation, and large - fluctuation. The range of the new - energy output volatility for small - fluctuation is [R th1 , R th2 ), the range of the new - energy output volatility for medium - fluctuation is [R th2 , R th3 ), and the range of the new - energy output volatility for large - fluctuation is [R th3 , +∞), where R th1 , R th2 , and R th3 are the preset first volatility threshold, second volatility threshold, and third volatility threshold, respectively; According to the duration T of the new energy output fluctuation event dur it is divided into short-term fluctuations and long-term fluctuations. The duration range of the short-term fluctuations is (0, T th ), and the duration range of the long-term fluctuations is [T th , +∞), where T th is a preset time threshold; According to the positive or negative of the new energy output fluctuation amount \(\Delta p\), the fluctuation direction is divided into upward fluctuation and downward fluctuation.

3. The distributed self-healing restoration method for a power system considering the uncertainty of new energy output according to claim 2, characterized in that The specific method for collecting meteorological data and the operation data of new energy units, training a short-term prediction model for predicting new energy output, using the short-term prediction model to predict the new energy output at future times, and determining whether the trigger conditions for new energy output fluctuation events are met is as follows: Configure meteorological sensors and electrical sensors at the new energy station to collect meteorological data and the operation data of new energy units. The meteorological data includes historical meteorological data and weather forecast data, and the operation data includes the inverter output power and the total output of the station; Select the long short-term memory network as the initial network of the short-term prediction model, preset the sliding window length \(n_t\) and the sliding step size, and generate a time series from the historical meteorological data, operation data, and weather forecast data in chronological order; Take the historical meteorological data and operation data of the \(n_t\) moments before the \(t\)th moment and the weather forecast data of the \(n_t\) moments after the \(t\)th moment as the input data of the short-term prediction model, and take the future new energy output as the output data. The sliding window moves backward on the time series in turn according to the sliding step size, and generates training samples on the time series in the form of a sliding window; Define the mean square error as the loss function for training the long short-term memory network, use the training samples to train the long short-term memory network, optimize the model parameters by minimizing the value of the loss function, and obtain the trained short-term prediction model when the loss function converges. Every time window \(m_t\), the historical meteorological data and operation data of the \(n_t\) moments before the current moment and the weather forecast data of the \(n_t\) moments after are input into the short-term prediction model to predict the future new energy output. According to the predicted new energy output, the new energy output fluctuation amount \(\Delta p\) and the new energy output volatility \(R_a\) are calculated; Compare the new energy output volatility with the first volatility threshold R th1 If the new energy output volatility is greater than or equal to the first volatility threshold, a new energy output fluctuation event is triggered.

4. The distributed self-healing recovery method for a power system considering the uncertainty of new energy output according to claim 3, characterized in that, The specific method for evaluating the severity of the new energy output fluctuation event is as follows: Judge the severity of the fluctuation amplitude, duration, and fluctuation direction of the new energy output fluctuation event; Among them, the method for judging the fluctuation amplitude is: Define the severity of the fluctuation amplitude as 1, 2, and 3 for small fluctuations, medium fluctuations, and large fluctuations respectively. According to the magnitude of the new energy output volatility, judge the value of the severity \(R_s\) of the fluctuation amplitude, where \(R_s=\{1, 2, 3\}\); The method for judging the duration is as follows: obtain the new energy output sequence based on the new energy output predicted n times, and judge the moment t1 when the new energy output volatility Ra is greater than or equal to the first volatility threshold R for the first time as the starting moment of the new energy output fluctuation event; th1 ​ Starting from time t1, determine the time t2 when the volatility Ra of new energy output is less than the first volatility threshold R for the first time after that, and use it as the end time of the new energy output fluctuation event. Obtain the duration based on the difference between time t1 and time t2; th1 ​ Define the severity of the duration as 1 and 2 for short-term fluctuations and long-term fluctuations respectively. According to the magnitude of the duration, judge the value of the severity \(R_T\) of the duration, where \(R_T = \{1, 2\}\); Among them, the method for judging the fluctuation direction is: Define the severity of the fluctuation direction as 1 and 2 for upward fluctuations and downward fluctuations respectively. According to the positive and negative of the new energy output fluctuation amount, judge the value of the severity \(R_d\) of the fluctuation direction, where \(R_d=\{1, 2\}\); According to the severity of the fluctuation amplitude, duration, and fluctuation direction, calculate the severity \(S\) of the new energy output fluctuation event; Set the severity thresholds \(S_1\) and \(S_2\), and classify the severity of the new energy output fluctuation event into mild fluctuations, moderate fluctuations, and severe fluctuations; among them, the range of the severity of mild fluctuations is: \(1\leq S < S_1\), the range of the severity of moderate fluctuations is: \(S_1\leq S < S_2\), and the range of the severity of severe fluctuations is: \(S_2\leq S < 3\).

5. The distributed self-healing recovery method for a power system considering the uncertainty of new energy output according to claim 4, wherein, The specific method for designing the state, action, and reward function of the intelligent agent and optimizing the self-healing control strategy using the reinforcement learning algorithm is as follows: Define the state space of the agent. The state s includes the volatility Ra of new energy output, the grid frequency deviation Δf, the state of charge SOC of the energy storage system, the state of controllable loads, and the severity S of new energy output fluctuation events. Define the action space of the agent. The action a includes the new energy output restriction command P cur , the energy storage charge and discharge power command P ess , and the controllable load control command P load ; Based on the grid frequency deviation Δf, the new energy output limit |P cur |, the charge change of the energy storage system |ΔSOC|, the controllable load control amount |P load | and the effect evaluation value of historical actions to define the reward function of the agent; Using a reinforcement learning algorithm, initialize the parameters of the current Q-network and, at the state s at the current moment t select an action randomly with probability ε or select the action with the maximum Q-value with probability 1 - ε, and execute the selected action a t to obtain the state s at the next moment t+1 and the reward function value r t ; Store s t , a t , r t , s t+1 into the experience replay buffer, and randomly sample a batch of samples s j , a j , r j , s j+1 from the experience replay buffer, and calculate the target value of the current Q-network; Define the loss function \(L(\theta)\) of the current Q-network, update the parameters \(\theta\) of the current Q-network by gradient descent method, minimize the value of the loss function, and periodically update the target Q-network parameters \(\theta\) - to the parameters \(\theta\) of the current Q-network; The intelligent agent selects the optimal action according to the current state. When selecting an action, if an action outside the constraint range is selected, reselect until an action within the constraint range is selected, and an optimized self-healing control strategy is obtained.

6. The distributed self-healing recovery method for a power system considering the uncertainty of new energy output according to claim 5, characterized in that The specific method for establishing the self-healing control strategy for the new energy output fluctuation event is as follows: Set the self-healing control strategy according to the severity of the new energy output fluctuation event, which is used as the constraint range of the new energy output limit command P cur in the action space, the charge and discharge power command P ess of the energy storage system, and the controllable load control command P load ; among them, when the fluctuation is mild, the constraint range of the new energy output limit command is P cur ∈[0.9P0, P0], the constraint range of the charge and discharge power command of the energy storage system is and the constraint range of the controllable load control command is P load ∈[0.95P l , 1.05P l ; when the fluctuation is moderate, the constraint range of the new energy output limit command is P cur ∈[0.8P0, 0.9P0), the constraint range of the charge and discharge power command of the energy storage system is and the constraint range of the controllable load control command is P load ∈[0.9P l , 1.1P l ; when the fluctuation is severe, the constraint range of the new energy output limit command is P cur ∈[0.7P0, 0.8P0), the constraint range of the charge and discharge power command of the energy storage system is and the constraint range of the controllable load control command is P load ∈[0.85P l , 1.15P l , where P0 is the initial output of the new energy, is the rated power of the energy storage system, and P l is the initial power of the controllable load.

7. The distributed self-healing restoration method for a power system considering the uncertainty of new energy output according to claim 6, characterized in that The specific method for obtaining the effect evaluation value of the self-healing control of the new energy output fluctuation event and feeding the effect evaluation value back to the reinforcement learning algorithm to guide the optimization of the next round of self-healing control strategy is as follows: After implementing the optimized self-healing control strategy, obtain the effect evaluation indicators of self-healing control. The effect evaluation indicators include: the grid frequency recovery time T fr , the energy storage cycle life loss L ess and the controllable load regulation cost C load ; Calculate the effect evaluation value of the self-healing control of the new energy output fluctuation event according to the effect evaluation index; Feed the effect evaluation value obtained by executing the current self-healing control strategy back to the reinforcement learning algorithm to guide the optimization of the next self-healing control strategy.

8. A distributed self-healing restoration system for a power system considering the uncertainty of new energy output, which is implemented based on the distributed self-healing restoration method for a power system considering the uncertainty of new energy output according to any one of claims 1-7, characterized in that, Including: The fluctuation event classification module is used to define the new energy output fluctuation amount and the new energy output volatility, set the trigger condition of the new energy output fluctuation event, and classify the new energy output fluctuation event according to the fluctuation amplitude, duration, and fluctuation direction; A severity calculation module, which is used to collect meteorological data and the operation data of new energy units, train a short-term prediction model for predicting new energy output, use the short-term prediction model to predict the new energy output at future times, determine whether the triggering conditions for new energy output fluctuation events are met, and if so, evaluate the severity of the new energy output fluctuation events; A self-healing strategy optimization module, which is used to establish a self-healing control strategy for new energy output fluctuation events, design the state, actions and reward functions of the intelligent agent, and optimize the self-healing control strategy by using a reinforcement learning algorithm; A self-healing strategy execution module, which is used for each control unit of the power system to monitor new energy output fluctuation events in real time. When a new energy output fluctuation event is triggered, the optimized self-healing control strategy is sent to the distributed control unit of the power system to execute the optimized self-healing control strategy; An effect evaluation feedback module, which is used to obtain the effect evaluation value of the self-healing control of new energy output fluctuation events, feedback the effect evaluation value to the reinforcement learning algorithm, and guide the optimization of the next round of self-healing control strategies.

9. An electronic device, characterized in that, It includes a memory, a processor and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in the power system distributed self-healing recovery method considering the uncertainty of new energy output as described in any one of claims 1-7 are implemented.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program, and the computer program is suitable for being loaded by the processor to execute the steps in the power system distributed self-healing recovery method considering the uncertainty of new energy output as described in any one of claims 1-7.

Images