Power system network key node identification method, power system network key node recovery optimization method and power system network key node identification device

Abstract

The invention discloses a power system network key node identification method, recovery optimization method and device, and the method comprises the steps: constructing a DQN model to learn the motion of selecting an optimal removal node in different network states, and enabling the motion in the model to remove nodes in a network and corresponding ring layer nodes; training a DQN model by using a training data set comprising power system network information in different states, generating a damaged key node identification model, and constructing a reward function by using an influence proportion of a network path after nodes and circle layer nodes are removed, a proportion of affected nodes and a connectivity value of the influence path in a training process; and obtaining operation data of a target power system network, constructing a network topology structure and a distance matrix, and inputting the network topology structure and the distance matrix into the damaged key node identification model to obtain an identification result of the damaged key node. The method has the advantages of being simple in implementation method, low in implementation cost and complexity, high in execution efficiency and precision, high in generalization ability and robustness and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of power system security monitoring, and in particular to a power system network key node identification method, recovery optimization method and device. Background Art

[0002] During the operation of the power system network, when it is affected by emergencies such as natural disasters, equipment failures, and human attacks, nodes may be damaged, affecting the normal operation of the power grid. If the damaged key nodes in the disaster scenario can be accurately identified, it is easy to restore and optimize the damaged nodes to ensure the stability and reliability of the power grid. For the identification of key nodes, the existing technology usually adopts methods based on complex network theory or electrical parameters. The method based on complex network theory is to identify key nodes in the power grid by calculating the topological properties of the nodes (such as node degree, betweenness centrality, etc.), and the method based on electrical parameters is to monitor the electrical parameters of each node in the power grid (such as voltage, current, power, etc.), analyze the parameter changes of the nodes before and after the failure, and thus identify the damaged nodes. However, the above-mentioned methods based on complex network theory or electrical parameters have high computational complexity and low efficiency, and are difficult to apply to the current power grid environment with ever-expanding scale. Moreover, it can only identify more important nodes based on the node attributes in the power grid, and it is impossible to quantify the impact of disasters on the connectivity of the power grid, so it is difficult to determine which are the key nodes affected by actual disasters.

[0003] The use of machine learning and deep learning technologies to automatically identify damaged nodes, that is, automatically identifying damaged nodes by analyzing images, videos or text data of power grid equipment, can improve the efficiency of identification and the degree of automation. However, the complexity of the traditional machine learning model is still relatively high and the generalization ability is insufficient, making it difficult to adapt to different power grid structures and operating conditions. The power grid structure and operating status are dynamically changing, and the traditional recognition method based on the machine learning model is difficult to adapt to the changing power grid structure and operating status, resulting in the actual recognition accuracy is still low, and the stability under different operating conditions is insufficient, and it is still difficult to meet the high reliability and high stability requirements of power grid operation. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in response to the problems existing in the prior art, the present invention provides a method for identifying key nodes in a power system network and a recovery optimization method and device, which have a simple implementation method, low implementation cost and complexity, high execution efficiency and accuracy, and strong generalization and robustness. The method can quickly and accurately identify key damaged nodes in the power grid under different disaster conditions, and then determine an effective node recovery optimization strategy.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is: A method for identifying key nodes in a power system network, comprising the steps of: A DQN (Deep Q-Network) model is constructed to learn the action of selecting the best node to remove under different network states, including the normal network connectivity state and the network state under different disaster types. The input feature of the model is the state representation of the power system network. The output of the model is a set of node sequences arranged in the order of removal and used as the identified damaged key nodes. The action of the intelligent agent in the model is to remove the nodes in the network and the corresponding circle nodes. The circle nodes are nodes within a specified distance centered on the target node. The DQN model is trained using a training data set consisting of power system network information under different network states to generate a damaged key node identification model, and a reward function is constructed using a node removal impact index during the training process, wherein the node removal impact index includes the impact ratio of the network path after removing the node and the circle node, the affected node ratio, and the connectivity value of the impact path; Acquire the operation data of the target power system network, and construct a network topology structure and a distance matrix according to the operation data of the target power system network, wherein each element in the distance matrix corresponds to the shortest path distance between nodes in the network; The operation data, topological structure and distance matrix of the target power system network are input into the trained damaged key node identification model to obtain the identification results of the damaged key nodes in the current target power system network.

[0006] Furthermore, the constructing of a network topology structure and a distance matrix according to the operation data of the target power system network includes: Collecting operation data of the power system network, including voltage level, current capacity of the connection, and status of the transmission line; According to the collected operation data of the power system network, the node and edge information in the power system network is identified, and the relationship between the nodes and edges is obtained, the connection mode and path between the nodes in the power grid are determined, and the topological structure diagram of the power system network is constructed; Computation graph based on node and edge information in power system network The adjacency matrix of ,in is the collection of nodes in the power system network. is the set of edges in the power system network, Representation Node To Node The weight of the edge; According to the adjacency matrix The shortest distance between any pair of nodes is obtained to obtain the distance matrix.

[0007] Furthermore, the adjacency matrix Configure as follows:

[0008] Distance Matrix The elements in are determined according to the following formula:

[0009]

[0010] in, Represented by adjacency matrix ,node and nodes Execute Dijkstra's algorithm as input, Represents a slave node To Node The shortest path distance.

[0011] Furthermore, the calculation expression of the constructed reward function is:

[0012] in, and They represent the weight parameters for measuring the proportion of affected paths and the proportion of affected nodes, represents the connectivity measure of the network after removing the node, Indicates the impact ratio of the network path after removing nodes and circle nodes. Indicates the proportion of nodes affected after removing nodes and circle nodes. Indicates the connectivity value affected by removing nodes and circle nodes; Impact ratio of network paths after removing nodes and circle nodes The calculation expression is:

[0013] in, Represents the number of nodes in the remaining network after removing nodes and circle nodes. The number of nodes that a node can reach, represents the number of reachable node pairs in the power system network, Represents the total number of nodes in the network, Represents the node set of the remaining network after removing nodes and circle nodes; The proportion of nodes affected after removing nodes and circle nodes The calculation expression is:

[0014] in Indicates the number of circle nodes, Represents the shortest distance centered on the removed node The circle node set formed by the nodes within; Connectivity values ​​affected by removing nodes and circle nodes The calculation expression is:

[0015] in, Indicates the removal of a node and the maximum connected subgraph after the corresponding circle node The number of nodes, Represents the set of nodes to be removed. Represents the largest connected subgraph of the power system network The number of nodes.

[0016] Further, training the DQN model using a training data set comprising power system network information in different states includes: The training data set containing the power system network information in different states is used as the input of the DQN model; During the training process, each step evaluates all possible removal actions based on the current network state, that is, the nodes to be removed and the corresponding circle nodes. As the current removed node, the DQN model predicts the reward according to the reward function and finds the node with the maximum reward , then remove the node set Then select the next node as the current removal node and determine the node with the maximum reward , then remove the node set , and so on, using a gradual expansion strategy, in the The node with the largest reward is predicted , find a set of node removal sequences , Indicates the number of removed nodes; Determine the set of nodes to be removed Whether it contains all the nodes of the original network, that is, Is it equal to If yes, the entire network is considered to be completely destroyed, the training stops, and a set of node sequences arranged in the order of removal is obtained as the identified damaged key nodes, otherwise the model continues to be trained. The removed node set For a node collection And the corresponding circle node set The union of .

[0017] A power system network disaster simulation and restoration optimization method, the steps comprising: According to the above method, the damaged key node set in the target power system network is identified , L Indicates the scope of the damaged node; From the set of damaged key nodes identified A set of nodes is randomly selected from And the corresponding circle nodes Interrupt to simulate different disaster scenarios, and get the current set of damaged nodes as , For a node collection And the corresponding circle node set The key node set to be repaired first is ; Obtain the network status of the target power system network under the current disaster scenario simulation state and the number of connected subgraphs; According to the network status under the current disaster scenario simulation state and the number of connected subgraphs, nodes are selected for restoration and the backup lines between the selected nodes and the remaining network nodes are enabled, and the damaged key nodes in the network are restored first. If the network is in a disconnected state, the nodes that are fastest connected to each connected subgraph are identified and used as target repair nodes; the target repair nodes are repaired first until the network forms a single connected subgraph so that the network reaches a connected state; when the network reaches a connected state, the target nodes that have the best network performance are selected in turn for restoration and the backup lines of the target nodes are enabled to obtain the optimal repair node sequence.

[0018] Furthermore, a hierarchical reinforcement learning model is constructed and used to perform node recovery, wherein the underlying network in the hierarchical reinforcement learning model preferentially repairs the target repair node until the network forms a single connected subgraph so that the network reaches a connected state, and the state of the underlying network is defined as the number and size distribution of the remaining network connected subgraphs, and the action is defined as selecting a node for repair and enabling the backup line of the node to be repaired; the high-level network in the hierarchical reinforcement learning model is used to sequentially select the target nodes that optimize the network capacity for recovery, and the state of the high-level network is defined as the current remaining network and the set of nodes that need to be repaired and the connectivity measure of the entire network, and the action is defined as selecting a node that improves network performance for repair and enabling the backup line of the selected node.

[0019] Furthermore, the reward function of the underlying network is defined as , represents the maximum connected subgraph size of the network after the node is repaired, represents the number of connected subgraphs of the network after the node is repaired, and They represent the weight parameters for measuring the maximum size of the network's connected subgraph and the number of the network's connected subgraphs, respectively. Indicates the first repair node The average clustering coefficient of the posterior network, It is the weight to measure the priority of repairing key nodes. is the first indicator function, used when the first repair node Belongs to the priority repair key node set The value is 1 when it is, otherwise it is 0; The reward function of the high-level network is defined as , Indicates repair node The average clustering coefficient of the posterior network, Indicates that according to the second repair node The connectivity measure of the network is determined by the number of nodes in the largest connected subgraph in the remaining network except Indicates that except for the second repair node The remaining network, and represents the weight parameter, is the second indicator function, when the second repair node Belongs to the priority repair key node set The value is 1 when , otherwise it is 0.

[0020] Further, using the hierarchical reinforcement learning model to perform node recovery includes: Building the underlying state table and high-level state table It is used to store the expected return value of each state-action pair of the underlying network and the high-level network; Obtaining a set of nodes removed from the power system network under different disaster types and remaining network data and inputting them into the hierarchical reinforcement learning model; In the hierarchical reinforcement learning model, the underlying network is in each state Next, according to the underlying state table and exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , Represents the discount factor; repeat the process until the number of connected subgraphs of the network becomes 1 after the node is repaired, and then switch to the high-level network for node recovery; When a high-level network performs node recovery, in each state According to the high-level status table And the exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , repeat until the preset iteration stop condition is reached to determine the optimal underlying state table and high-level state table , and obtain a hierarchical reinforcement learning model with optimal parameters; The node set removed from the current power system network and the remaining network data are obtained and input into the hierarchical reinforcement learning model with optimal parameters to perform node recovery.

[0021] A power system network disaster simulation recovery optimization device comprises a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0022] Compared with the prior art, the advantages of the present invention are: 1. The present invention combines the quantification of disaster impacts with a deep reinforcement learning method to construct a DQN model to learn to select the best action to remove nodes under different network states. Each action performed by the intelligent agent in the model includes removing nodes in the network and corresponding circle nodes. The impact of removed nodes on nodes in the circle can be comprehensively considered, and a reward function is constructed using the node removal impact index during model training. The impact of removed nodes on grid connectivity can be accurately quantified, and the impact of different disasters on grid connectivity can be accurately evaluated. The deep reinforcement learning algorithm is used to dynamically learn and optimize the nodes in the grid, which can quickly and accurately identify damaged key nodes affected by different disasters, and the recognition accuracy is not easily affected by changes in grid structure and operating status, which can improve the generalization ability of the model and the robustness of recognition.

[0023] 2. The present invention can further simulate different disaster scenarios and evaluate the damage to key nodes and paths to simulate the impact of natural disasters or other emergencies on the power grid. At the same time, it can use reinforcement learning strategy methods to guide the node repair work of the power grid, determine the node repair priority according to the network status of whether the network is connected, and perform node recovery in a staged manner. First, the target repair node is repaired first and its backup line is enabled until the network forms a single connected subgraph so that the network reaches a connected state. This can accurately realize the rapid repair of the power network after a disaster. Then, when the network reaches a connected state, the target node with the best network performance is selected in turn for recovery and its backup line is enabled, so that not only the rapid recovery of the network can be achieved, but also the optimization of the network performance after recovery can be effectively achieved, ensuring the rapid response and efficient recovery of the power grid in the face of disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the implementation flow of the method for identifying key nodes in the power system network according to embodiment 1 of the present invention.

[0025] Figure 2 A schematic diagram of the architecture of the DQN model used in Example 1 of the present invention.

[0026] Figure 3 It is a schematic diagram of the implementation process of node recovery using a hierarchical reinforcement learning model in Example 2 of the present invention.

[0027] Figure 4 It is a schematic diagram of the principle of realizing key node identification and node recovery in a specific application embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0029] The present invention combines the quantification of disaster impact and deep reinforcement learning methods, introduces the impact ratio of the network path after removing nodes and circle nodes, the ratio of affected nodes and the connectivity value of the impact path as the node removal impact index, so as to quantify the impact state after removing nodes in different network states, and constructs a DQN model to learn to select the best node removal action under different network states. The action performed by the intelligent agent in the model each time includes removing nodes in the network and the corresponding circle nodes, which can comprehensively consider the impact of removing nodes on nodes in the circle, and use the node removal impact index to construct a reward function during the model training process, so that the action is rewarded or punished according to the impact state of the removed node, and the impact of the grid connectivity after the node is removed can be accurately quantified, and the impact of different disasters on the grid connectivity can be accurately evaluated, so that the nodes in the grid are dynamically learned and optimized by using the deep reinforcement learning algorithm, and the damaged key nodes affected by different disasters can be quickly and accurately identified, and the recognition accuracy is not easily affected by changes in the grid structure and operating status, which can improve the generalization ability of the model and the robustness of recognition.

[0030] like Figure 1 As shown, the steps of the method for identifying key nodes in the power system network of this embodiment include: Step S01. Construct a DQN model to learn to select the best action to remove nodes under different network states. The network state includes the normal network connectivity state and the network state under different disaster types, where the disaster types include typhoons, rainstorms, floods and other types of extreme weather disasters. The input feature of the model is the state representation of the power system network, which includes information such as nodes and edges in the network. The output of the model is a set of node sequences arranged in the order of removal and used as the identified damaged key nodes. The action of the intelligent agent in the model is to remove the nodes in the network and the corresponding circle nodes. The circle nodes are nodes within a specified distance centered on the target node.

[0031] In order to quantify the impact of node removal on surrounding nodes and the performance of the entire network, this embodiment defines an impact circle to simulate the impact of node removal on circle nodes and the performance of the entire network. Specifically, according to the impact of disasters of different degrees, the scope of the affected nodes caused by the disaster is defined as , remove any node in the network , then the nodes in the disaster impact circle are defined as ,in Represents a slave node To Node For example, when When the value is 1, the circle node is the node The direct neighbor node of When the value is other than , the node set can be obtained according to the Dijkstra algorithm The shortest distance from each node to other nodes in the The nodes within are circle nodes.

[0032] In order to construct a DQN model to learn to select the best action to remove nodes under different network conditions, in a specific application embodiment, such as Figure 2 As shown, the deep reinforcement learning model can be defined as follows: Step S101: Initialize the current value network and the target value network, set the number of iterations C and various weights, and initialize the experience playback buffer D (experience pool).

[0033] Step S102: Obtain the power system network information under different network states to form a training data set, and output The power system network information includes network nodes, edges and other information. For example, an undirected network with 15 nodes and 24 edges can be used as the input of the model.

[0034] Step S103. Use Select the action a to be performed in this way, that is, Probability selects the action with the highest Q value in the current state, The probability of randomly selecting actions can be expressed as: , (5) in, is the selected action, is the current state, is the exploration probability, represents the number of current trainings, M represents the total number of trainings, It is the action with the highest Q value in the current state. Indicates random selection.

[0035] As training progresses, The value of will continue to decrease. At this time, the maximum Q value is used with a greater probability to select actions, which can speed up the training process.

[0036] Step S104: Select an action Then get the reward R and get the next state .

[0037] Step S105. Store them in the experience replay buffer D. If the experience replay buffer D is full, delete the old samples and store the new samples.

[0038] Step S106. Randomly extract a set of samples from the experience replay buffer D and use the gradient descent method to update the weights of the DQN network: (6) in, is the learning rate, is the gradient of the current network parameters.

[0039] Step S107. After multiple rounds of iterations, the weights of the DQN network are updated until the maximum number of iterations is reached or other preset iteration conditions are met, and the optimal model parameters are saved to obtain the trained model. The following minimization loss function can be specifically used in the iteration process: (7) in, are the parameters of the current network.

[0040] Step S02. Use a training data set containing network information of the power system under different network states to train the DQN model, generate a damaged key node identification model, and use the node removal impact index to construct a reward function during the training process. The node removal impact index includes the impact ratio of the network path after removing the node and the circle node, the proportion of affected nodes, and the connectivity value of the affected path. The impact ratio of the network path after removing the node and the circle node is used to characterize the proportion of the network path affected after removing the node and the circle node, the proportion of affected nodes is used to characterize the proportion of nodes affected after removing the node and the circle node, and the connectivity value of the affected path is used to characterize the impact of removing the node and the circle node on the connectivity of the network path.

[0041] This embodiment combines graph theory analysis and deep reinforcement learning models, trains the DQN model to learn the best action of selecting nodes to remove under different network states, uses the model to analyze the operation data of the power grid, and defines the reward function using the removal node impact index to identify the key node sequence that is critical to the operation of the power grid. The impact of the lost nodes on the stability and power supply capacity of the entire power grid can be determined. These nodes play a core role in the power grid, and their failure or damage may have a significant impact on the power grid, thereby realizing the identification of damaged key nodes.

[0042] In order to quantify the index of the impact of disasters on the network, the impact index of each removed node after removing the node and circle node in this embodiment is specifically defined according to the following steps: (1) Impact ratio of network paths When several nodes are removed from the network, the set of damaged nodes is defined as , represents the number of removed nodes, then the circle node union of the disaster-affected node set is defined as: ,in Traversal All nodes in , the removed node set is defined as , after removing the nodes and circle nodes, the node set of the remaining network is .

[0043] Assume that the number of reachable node pairs in the original network is , Represents the total number of nodes in the network, then for each node in the remaining network , Dijkstra algorithm can be used to calculate other nodes The shortest path and record each The number of nodes that can be reached (excluding itself), and the number of reachable node pairs in the remaining network is , then as an optional implementation, the impact ratio of the network path after removing nodes and circle nodes can be defined for: (1) in, Represents the number of nodes in the remaining network after removing nodes and circle nodes. The number of nodes that a node can reach, A collection of nodes representing the residual network.

[0044] (2) Proportion of affected nodes As an optional implementation, the proportion of nodes affected after removing nodes and circle nodes It can be defined as the ratio of the number of circle nodes in a node set to the total number of nodes in the network. For example, the calculation expression can be expressed as: (2) in, Indicates the number of circle nodes, Represents the shortest distance centered on the removed node The circle node set formed by the nodes within.

[0045] (3) Impacted connectivity value As an optional implementation, the affected connectivity value can be defined as the ratio of the number of nodes in the largest connected subgraph of the original network minus the number of nodes in the largest connected subgraph of the remaining network to the number of nodes in the largest connected subgraph of the original network, for example The calculation expression can be expressed as: (3) in, Indicates the removal of a node and the maximum connected subgraph after the corresponding circle node The number of nodes, Indicates the removal of a node The node set after Represents the largest connected subgraph of the power system network The number of nodes, It is a measure of the connectivity of the network after removing the node, which can be specifically defined as the number of nodes in the largest connected subgraph in the remaining network.

[0046] This embodiment introduces circle nodes and node removal impact indicators to form a comprehensive evaluation index system for the impact of removing nodes from the network. This system is used to analyze the collection of nodes in the network and their circles, and can accurately quantify the impact of different disaster events on grid connectivity, breaking through the limitations of traditional single-node evaluation, and effectively evaluating the impact of nodes and edges on the power network in disaster scenarios, thereby enabling a more comprehensive identification of key nodes.

[0047] As an optional implementation, this embodiment considers the change in network connectivity after removing nodes and circle nodes for the node removal sequence, and uses the above-mentioned impact indicators of each removed node to construct a reward function as follows: (4) in, and They represent the weight parameters for measuring the proportion of affected paths and the proportion of affected nodes, represents the connectivity measure of the network after removing the node, represents the influence ratio of the path, Indicates the proportion of nodes affected, The connectivity value representing the influence.

[0048] The reward function constructed in the above manner can use the impact ratio of the network path and the proportion of the affected nodes to measure the degree to which a node is affected by a disaster and damages the entire network. At the same time, considering the connectivity measurement, it can determine whether the network is disconnected after each node is affected by the disaster. The reward function is defined by comprehensively removing the impact ratio of the network path after the nodes and circle nodes, the proportion of the affected nodes, and the connectivity value of the affected path. It can prioritize identifying nodes that have a high degree of damage to the network and cause the network to be disconnected into several connected subgraphs, thereby accurately identifying damaged key nodes.

[0049] This embodiment trains the DQN model and finally outputs a set of critically ranked node sequences found by the reinforcement learning model. The nodes in the node sequence are arranged in the order of removal, and the arrangement order can represent the impact of each node on the destructiveness of the network. As an optional implementation, training the DQN model using the training data set includes: Step S201. Using a training data set consisting of power system network information in different states as input to the DQN model; Step S202. During the training process, each step evaluates all possible removal actions based on the current network status, that is, the nodes to be removed and the corresponding circle nodes, by first selecting a node As the current removed node, the DQN model predicts the reward according to the reward function and finds the node with the maximum reward , then remove the node set Then select the next node as the current removal node and determine the node with the maximum reward , then remove the node set , and so on, using a gradual expansion strategy, The set of removed nodes is , in The node with the largest reward is predicted , find a set of node removal sequences ; Step S203: Determine the node set to be removed Whether it contains all the nodes of the original network, that is, , node set and circle node set If yes, the entire network is considered to be completely destroyed, and the training stops. A set of node sequences arranged in the order of removal is obtained. As the damaged key nodes identified, their node sorting order is sorted according to the comprehensive impact on the network, among which the set of removed nodes The node set And the corresponding circle node set The union of .

[0050] Step S03: Acquire the operation data of the target power system network, and construct a network topology structure and a distance matrix according to the operation data of the target power system network. Each element in the distance matrix corresponds to the shortest path distance between nodes in the network.

[0051] As an optional implementation, constructing a network topology structure and a distance matrix according to the operation data of the target power system network includes: Step S301. Collect operation data of the power system network, the operation data including voltage level, current capacity of the connection, status of the transmission line, etc.

[0052] Optionally, after collecting the operation data of the power grid, the collected data is also cleaned and preprocessed to ensure the accuracy and availability of the data.

[0053] Step S302. Identify the nodes and edge information in the power system network based on the collected operation data of the power system network, obtain the relationship between the nodes and edges, determine the connection mode and path between each node in the power grid, and construct a topological structure diagram of the power system network.

[0054] By analyzing the operation data of the power system network and sorting out the node and edge information of the power grid, we can construct a topological structure diagram of the power grid. By analyzing the relationship between each node and edge, we can determine the connection method and path between each node in the power grid.

[0055] Step S303: Calculate the graph based on the node and edge information in the power system network The adjacency matrix of ,in is the collection of nodes in the power system network. is the set of edges in the power system network, Representation Node To Node The edge weights are calculated based on the adjacency matrix Get the shortest distance between any pair of nodes and get the distance matrix .

[0056] The adjacency matrix is ​​used to represent the connection relationship between nodes in the power grid, where the elements of the matrix represent the transmission capacity or weight between nodes. It is constructed using the Dijkstra algorithm, that is, the Dijkstra algorithm can be used with the help of the adjacency matrix Calculate and find the shortest path between any pair of nodes in the power grid, thus obtaining the distance matrix Specifically, a graph is constructed based on the node and edge information of the power grid. , determine the total number of nodes in the network and the total number of edges , for each edge Assign weight , the weight represents the transmission capacity of the edge.

[0057] As an optional implementation, the edge weights can be set to 1, indicating that the transmission capacity of each edge of the power grid is similar, that is, Representation Node To Node The weight of the edge, if the node and nodes If there is no edge between , indicating unreachable, and the diagonal elements , adjacency matrix Configure as follows: (8) Then use Dijkstra's algorithm based on the adjacency matrix to find the distance between any pair of nodes. Specifically, for each node , run Dijkstra's algorithm to find To other nodes The shortest path distance : , and get the distance matrix , that is, the distance matrix The elements in are determined according to the following formula: (9)

[0058] in, Represented by adjacency matrix ,node and nodes Execute Dijkstra's algorithm as input, Represents a slave node To Node The shortest path distance.

[0059] Step S04: Input the operation data, topological structure and distance matrix of the target power system network into the trained damaged key node identification model to obtain the identification result of the damaged key node.

[0060] This embodiment utilizes the damaged key node identification model obtained by training the DQN model in step S01. When it is necessary to identify the key nodes of the target power system network, the operation data of the target power system network is first obtained, and the network topology structure and distance matrix are constructed by analyzing the operation data, and then input into the trained damaged key node identification model. The model can quickly and accurately identify the damaged key nodes of the target power system network.

[0061] Through the above steps, this embodiment can combine the quantification of the impact of disasters on nodes and the deep reinforcement learning method, conduct an impact assessment on the circle nodes of the power grid nodes according to the impact of the disaster, and use the deep reinforcement learning method to dynamically learn and optimize the nodes in the power grid, so as to quickly and accurately identify the damaged key nodes of the network under different disasters, and improve the recognition efficiency, reliability and robustness of the key nodes.

[0062] Embodiment 2: This embodiment is based on Embodiment 1. By simulating different disaster scenarios and evaluating the damage to key nodes and paths, the impact of natural disasters or other emergencies on the power grid is simulated. At the same time, the reinforcement learning strategy method is used to guide the node repair work of the power grid. The node repair priority is determined according to the network status of whether the network is connected. The node recovery is performed in a staged manner. The target repair node is repaired first and its backup line is enabled until the network forms a single connected subgraph so that the network reaches a connected state. The rapid repair of the power network after a disaster can be accurately achieved. Then, when the network reaches a connected state, the target nodes with the best network performance are selected in turn for recovery and their backup lines are enabled. This not only enables the rapid recovery of the network and reduces the number of connected subgraphs in the network, but also effectively optimizes the network performance after recovery, ensuring the rapid response and efficient recovery of the power grid in the face of disasters.

[0063] The steps of the power system network disaster simulation and restoration optimization method in this embodiment include: Step 1: Identify the damaged key node set in the target power system network according to Example 1 , L Indicates the scope of the damaged node; Step 2: From the identified damaged key node set A set of nodes is randomly selected from And the corresponding circle nodes Interrupt to simulate different disaster scenarios, and get the current set of damaged nodes as , For a node collection And the corresponding circle node set The key node set to be repaired first is ; Step 3: Obtain the network status of the target power system network under the current disaster scenario simulation state and the number of connected subgraphs; Step 4: Select nodes for restoration based on the network status and the number of connected subgraphs under the current disaster scenario simulation state and enable the backup lines between the selected nodes and the remaining network nodes, giving priority to restoring damaged key nodes in the network. If the network is disconnected, identify the nodes that are fastest connected to each connected subgraph and use them as target repair nodes; prioritize repairing target repair nodes and enable their backup lines until the network forms a single connected subgraph so that the network reaches a connected state; when the network reaches a connected state, select the target nodes that have the best network performance in turn for restoration and enable the backup lines of the target nodes to obtain the optimal repair node sequence.

[0064] Specifically, according to the impact of disasters of different degrees, the scope of the damaged nodes caused by disasters is defined as , the set of nodes damaged by the disaster (damaged nodes) , then the node set The circle node union was also removed, among which Traversal All nodes in the node set can be calculated according to the Dijkstra algorithm The shortest distance from each node to other nodes in the network can then be determined to meet the shortest distance centered on the node. The nodes within are circle nodes, and the set of nodes to be removed can be determined as , the node set of the remaining network is , the set of nodes to be repaired is also .

[0065] In a specific application embodiment, in order to realize the optimization of power system network disaster simulation recovery, first analyze the post-disaster state of the power grid to determine whether the network is disconnected and the number of connected subgraphs; if the network is disconnected, identify the nodes that can connect to each connected subgraph the fastest, give priority to repairing these nodes that connect to each connected subgraph the fastest and enable their backup lines to reduce the number of connected subgraphs, and then continue the repair operation until the network forms a single connected subgraph; at this time, the network is connected, and the optimal node is selected from the remaining removed node set for recovery, and the node that can maximize the network capacity is selected for recovery and the backup line of the node to be restored is enabled, and then based on the current network state, the next optimal node is selected for recovery, and the selection and recovery process is repeated until the optimal repair node sequence is obtained, and the post-disaster recovery of the power grid is completed, ensuring that the connectivity and various capabilities of the power grid are maximized. The restored power grid can be further tested and evaluated to ensure stability and reliability.

[0066] As an optional implementation, a hierarchical reinforcement learning model can be constructed and used for node recovery, wherein the underlying network in the hierarchical reinforcement learning model prioritizes repairing the target repair node until the network forms a single connected subgraph so that the network reaches a connected state, and the state of the underlying network Defined as the number and size distribution of the remaining network connected subgraphs, actions It is defined as selecting a node for repair and enabling the backup line between the selected node and the remaining network nodes. The high-level network in the hierarchical reinforcement learning model is used to sequentially select the target node that optimizes the network capacity for recovery. The state of the high-level network Defined as the current remaining network and the set of nodes that need to be repaired as well as the connectivity measure of the entire network, action It is defined as selecting a node that improves network performance for repair and enabling the backup line between the selected node and the remaining network nodes. For example, the selection of this node is based on its increase in the network clustering coefficient and the optimization of the entire network performance. That is, a layered reinforcement method is adopted to connect the entire network from the bottom layer, that is, to merge the connected subgraphs of the remaining network, reduce the number of connected subgraphs in the network, and give priority to repairing those nodes that can connect multiple connected subgraphs. After repairing several nodes, the number of network connected subgraphs will be reduced. When the number of connected subgraphs in the repair process becomes 1, the strategy of focusing on merging connected subgraphs is automatically changed to a strategy of optimizing network performance, realizing dynamic hierarchical adjustment, and then using high-level optimization to optimize the connectivity and clustering coefficient of the network. After the number of connected subgraphs in the network becomes 1 during the repair process, the improvement of network connectivity and the increase of clustering coefficient are considered, which can effectively optimize the performance of the network after recovery and improve the efficiency and effect of repair.

[0067] Furthermore, the reward function of the underlying network can be defined as , represents the maximum connected subgraph size of the network after the node is repaired, represents the number of connected subgraphs of the network after the node is repaired, and They represent the weight parameters for measuring the maximum size of the network's connected subgraph and the number of the network's connected subgraphs, respectively. Indicates the first repair node The average clustering coefficient of the posterior network, It is the weight to measure the priority of repairing key nodes. is the first indicator function, used when the first repair node Belongs to the priority repair key node set The value is 1 when , otherwise it is 0; the reward function of the high-level network can be defined as , Indicates repair node The average clustering coefficient of the posterior network, Indicates that according to the second repair node The connectivity measure of the network is determined by the number of nodes in the largest connected subgraph in the remaining network except Indicates that except for the second repair node The remaining network, and represents the weight parameter, is the second indicator function, when the second repair node Belongs to the priority repair key node set The value is 1 when , otherwise it is 0.

[0068] In order to calculate the average clustering coefficient , , you can use the following methods: The average clustering coefficient is the clustering coefficient of the network, which is an indicator of the local connectivity of the network and can describe the degree of interconnection between the neighboring nodes of a node. A high clustering coefficient indicates that a node's neighboring nodes form a close community or group.

[0069] For nodes in the network , and its clustering coefficient can be expressed as ,in, Is a node The number of edges that actually exist between neighbor nodes, Is a node The degree of , then for the remaining network, the average clustering coefficient can be expressed as: ,in, is the node to be repaired. A collection of nodes representing the remaining network.

[0070] As an optional implementation, Figure 3 As shown, node recovery using a hierarchical reinforcement learning model includes: Step 401. Create a bottom-level status table and high-level state table It is used to store the expected return value of each state-action pair of the underlying network and the high-level network.

[0071] During the initial setup of the model, including initializing the underlying state table and high-level state table , set the learning rate and discount factor And the exploration rate , obtain the set of nodes removed from the original network and the remaining network data as input data.

[0072] Step 402: Obtain the set of nodes removed from the power system network under different disaster types and the remaining network data and input them into the hierarchical reinforcement learning model.

[0073] Step 403. In each state of the underlying network in the hierarchical reinforcement learning model Next, according to the underlying state table and exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , Represents the discount factor; repeat until the number of connected subgraphs of the network becomes 1 after the node is repaired, and then switch to the high-level network for node recovery.

[0074] Specifically, in each state, according to Table and Exploration Rate Choose actions based on probability choose The action with the largest value in the table, with probability Randomly select actions, that is: ,

[0075] in, is the selected action, is the current state, is the exploration probability, m represents the number of current trainings, and M represents the total number of trainings. is the action with the highest Q value in the current state. As training progresses, The value of will continue to decrease. At this time, the maximum Q value is used with a greater probability to select actions, which can speed up the training process.

[0076] Then perform the selected action , get rewards and the next state ,renew Table, using Q-learning to update the formula: renew Repeat the above steps. When the number of connected subgraphs of the network becomes 1 after the node is repaired, it will automatically transfer to the high-level logic to use Table and new reward function , that is, finding the repaired nodes based on the average clustering coefficient of the network and the connectivity measure of the network.

[0077] Step 404. When performing node recovery in the high-level network, in each state According to the high-level status table And the exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , repeat until the preset iteration stop condition is reached to determine the optimal underlying state table and high-level state table , and obtain a hierarchical reinforcement learning model with optimal parameters.

[0078] Specifically, according to Table and Exploration Rate Choose actions based on probability choose The action with the largest value in the table, with probability Randomly select an action and execute the selected action , get rewards and the next state ;renew Table, using Q-learning update formula: , repeat the above steps until the maximum number of iterations is reached or other stopping conditions are met. Finally, save the optimal and The table is used as the model parameters to obtain a hierarchical reinforcement learning model with optimal parameters.

[0079] Step 405: Obtain the set of nodes removed from the current power system network and the remaining network data and input them into a hierarchical reinforcement learning model with optimal parameters to perform node recovery.

[0080] In summary, this embodiment considers the impact of disasters on the entire power network. By constructing a removal strategy that includes circle nodes, combining deep reinforcement learning to identify damaged key nodes in the network, and then simulating different disaster situations, combining hierarchical reinforcement learning methods to generate node recovery strategies under different disaster situations, it is possible to accurately evaluate the impact of disasters on grid connectivity, identify key nodes, and optimize disaster recovery strategies. The hierarchical reinforcement learning strategy not only focuses on the rapid physical connection recovery of the network, but also takes into account the optimization of network performance after recovery, ensuring that the power grid can quickly restore normal power supply in the face of disasters, reducing the social and economic losses caused by power outages, thereby improving the power grid's ability to recover in disaster situations, and enhancing the overall robustness of the power grid, thereby significantly improving the power grid's response speed and recovery efficiency when disasters occur, enhancing the overall stability and robustness of the power grid, and facilitating power grid management and rapid disaster response, so that power grid management can be more refined, effectively preventing and reducing large-scale power outages, and ensuring the stable operation of the power grid.

[0081] To verify the effect of the present invention, the above method of the present invention is used in a specific application embodiment to identify key nodes in an undirected network with 15 nodes and 24 edges and to restore nodes under the influence of disasters. Figure 4As shown, firstly, the operation data of the network is obtained and analyzed to determine the nodes in the influence circle when each node is removed, and then the DQN model is used for learning to identify a group of critically ordered nodes as the identified node sequence, and then the optimal node recovery order is determined through hierarchical reinforcement learning. The serial number marked on the node side in the figure is the recovery serial number. The results can verify that the present invention can quickly and accurately identify damaged key nodes in the network, effectively determine the node recovery order, and optimize network performance.

[0082] To implement the methods of the above-mentioned embodiments 1 to 2, the device for identifying key nodes in a power system network of this embodiment includes: A model training module is used to construct a DQN model to learn the action of selecting the best node to remove under different network states, wherein the network state includes the normal network connectivity state and the network state under different disaster types, wherein the input feature of the model is the state representation of the power system network, and the output of the model is a set of node sequences arranged in the order of removal as the identified damaged key nodes, wherein the action of the intelligent agent in the model is to remove the nodes in the network and the corresponding circle nodes, and the DQN model is trained using a training data set consisting of power system network information under different network states to generate a damaged key node identification model, and a reward function is constructed using a node removal impact indicator during the training process, wherein the node removal impact indicator includes the impact ratio of the network path after removing the node and the circle node, the ratio of the affected nodes, and the connectivity value of the impact path, and the circle node is a node within a specified distance centered on the target node; The identification module is used to obtain the operating data of the target power system network, construct a network topology structure and a distance matrix according to the operating data of the target power system network, each element in the distance matrix corresponds to the shortest path distance between nodes in the network, and input the operating data, topology structure and distance matrix of the target power system network into the trained damaged key node identification model to obtain the identification result of the damaged key node in the current target power system network.

[0083] Furthermore, a power grid information collection and analysis module can be constructed to collect and analyze the operation data of the power grid, identify key information such as nodes, edges and transmission capacity in the power grid, so as to analyze the normal operation status of the power grid; and a power grid topology structure construction module can be constructed to construct the topology of the power grid by analyzing the edges and nodes of the power grid, and to construct the adjacency matrix of the power grid using methods such as complex network analysis; and a power grid distance matrix construction module can be constructed to calculate the shortest path distance between each node in the power grid, and then calculate the distance matrix of the shortest path between the nodes.

[0084] The device for identifying key nodes in a power system network of this embodiment corresponds one-to-one to the above-mentioned method for identifying key nodes in a power system network, and will not be described in detail here.

[0085] Furthermore, it also includes a power system network disaster simulation and restoration optimization device, including: The power system network key node identification device as described above; The disaster simulation module is used to randomly select a number of nodes and corresponding circle nodes from the identified damaged key nodes for interruption to simulate different disaster scenarios; A data acquisition module is used to obtain the network status of the target power system network and the number of connected subgraphs under the current disaster scenario simulation state; The node recovery module is used to select nodes for recovery according to the network status and the number of connected subgraphs under the current disaster scenario simulation state and enable the backup lines between the selected nodes and the remaining network nodes, giving priority to recovering damaged key nodes in the network. If the network is in a disconnected state, the node that is fastest connected to each connected subgraph is identified and used as the target repair node; the target repair node is repaired first until the network forms a single connected subgraph so that the network reaches a connected state; when the network reaches a connected state, the target nodes that optimize the network performance are selected in turn for recovery and the backup lines of the target nodes are enabled to obtain the optimal repair node sequence.

[0086] The power system network disaster simulation and recovery optimization device of this embodiment corresponds one to one with the above-mentioned power system network disaster simulation and recovery optimization method, and will not be described one by one here.

[0087] This embodiment further provides a computer device, including a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to perform the above method.

[0088] It is understandable that the above method of this embodiment can be executed by a single device, such as a computer or server, etc., and can also be applied to a distributed scenario and completed by multiple devices in cooperation with each other. In the case of a distributed scenario, one of the multiple devices can only execute one or more steps in the above method of this embodiment, and multiple devices interact to complete the above method. The processor can be implemented in the form of a general-purpose CPU, a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, etc., for executing related programs to implement the above method of this embodiment. The memory can be implemented in the form of a read-only memory ROM, a random access memory RAM, a static storage device, and a dynamic storage device. The memory can store an operating system and other applications. When the above method of this embodiment is implemented by software or firmware, the relevant program code is stored in the memory and called and executed by the processor.

[0089] Those skilled in the art should understand that the above-mentioned embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram and the combination of the processes and / or boxes in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0090] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment, it is not intended to limit the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for identifying key nodes in a power system network, characterized in that the steps include: A DQN model is constructed to learn the action of selecting the best node to remove under different network states, including the normal network connectivity state and the network state under different disaster types. The input feature of the model is the state representation of the power system network. The output of the model is a set of node sequences arranged in the order of removal and used as the identified damaged key nodes. The action of the intelligent agent in the model is to remove the nodes in the network and the corresponding circle nodes. The circle nodes are nodes within a specified distance centered on the target node. The DQN model is trained using a training data set consisting of power system network information under different network states to generate a damaged key node identification model, and a reward function is constructed using a node removal impact index during the training process, wherein the node removal impact index includes the impact ratio of the network path after removing the node and the circle node, the affected node ratio, and the connectivity value of the impact path; Acquire the operation data of the target power system network, and construct a network topology structure and a distance matrix according to the operation data of the target power system network, wherein each element in the distance matrix corresponds to the shortest path distance between nodes in the network; The operation data, topological structure and distance matrix of the target power system network are input into the trained damaged key node identification model to obtain the identification results of the damaged key nodes in the current target power system network.

2. The method for identifying key nodes in a power system network according to claim 1, characterized in that: The step of constructing a network topology structure and a distance matrix according to the operation data of the target power system network includes: Collecting operation data of the power system network, including voltage level, current capacity of the connection, and status of the transmission line; According to the collected operation data of the power system network, the node and edge information in the power system network is identified, and the relationship between the nodes and edges is obtained, the connection mode and path between the nodes in the power grid are determined, and the topological structure diagram of the power system network is constructed; Computation graph based on node and edge information in power system network The adjacency matrix of ,in is the collection of nodes in the power system network. is the set of edges in the power system network, Representation Node To Node The weight of the edge; according to the adjacency matrix The shortest distance between any pair of nodes is obtained to obtain the distance matrix.

3. The method for identifying key nodes in a power system network according to claim 2, characterized in that: Adjacency Matrix Configure as follows: Distance Matrix The elements in are determined according to the following formula: in, Represented by adjacency matrix ,node and nodes Execute Dijkstra's algorithm as input, Represents a slave node To Node The shortest path distance.

4. The method for identifying key nodes in a power system network according to claim 1, characterized in that: The calculation expression of the constructed reward function is: in, and They represent the weight parameters for measuring the proportion of affected paths and the proportion of affected nodes, represents the connectivity measure of the network after removing the node, Indicates the impact ratio of the network path after removing nodes and circle nodes. Indicates the proportion of nodes affected after removing nodes and circle nodes. Indicates the connectivity value affected by removing nodes and circle nodes; Impact ratio of network paths after removing nodes and circle nodes The calculation expression is: in, Represents the number of nodes in the remaining network after removing nodes and circle nodes. The number of nodes that a node can reach, represents the number of reachable node pairs in the power system network, Represents the total number of nodes in the network, Represents the node set of the remaining network after removing nodes and circle nodes; The proportion of nodes affected after removing nodes and circle nodes The calculation expression is: in Indicates the number of circle nodes, Represents the shortest distance centered on the removed node The circle node set formed by the nodes within; Connectivity values ​​affected by removing nodes and circle nodes The calculation expression is: in, Indicates the removal of a node and the maximum connected subgraph after the corresponding circle node The number of nodes, Represents the set of nodes to be removed. Represents the largest connected subgraph of the power system network The number of nodes.

5. The method for identifying key nodes in a power system network according to any one of claims 1 to 4, characterized in that: Training the DQN model using a training data set consisting of power system network information in different states includes: The training data set containing the power system network information in different states is used as the input of the DQN model; During the training process, each step evaluates all possible removal actions based on the current network state, that is, the nodes to be removed and the corresponding circle nodes. As the current removed node, the DQN model predicts the reward according to the reward function and finds the node with the maximum reward , then remove the node set Then select the next node as the current removal node and determine the node with the maximum reward , then remove the node set , and so on, using a gradual expansion strategy, in the The node with the largest reward is predicted , find a set of node removal sequences , Indicates the number of removed nodes; Determine the set of nodes to be removed Whether it contains all the nodes of the original network, that is, Is it equal to the set of nodes in the power system network? If yes, the entire network is judged to be completely destroyed, the training stops, and a set of node sequences arranged in the order of removal is obtained as the identified damaged key nodes. Otherwise, the model continues to be trained, and the set of removed nodes is For a node collection And the corresponding circle node set The union of .

6. A power system network disaster simulation and restoration optimization method, characterized in that the steps include: Identify a damaged key node set in the target power system network according to the method described in any one of claims 1 to 5 , Indicates the scope of the damaged node; From the set of damaged key nodes identified A set of nodes is randomly selected from And the corresponding circle nodes Interrupt to simulate different disaster scenarios, and get the current set of damaged nodes as , For a node collection And the corresponding circle node set The key node set to be repaired first is ; Obtain the network status of the target power system network under the current disaster scenario simulation state and the number of connected subgraphs; According to the network status under the current disaster scenario simulation state and the number of connected subgraphs, nodes are selected for restoration and the backup lines between the selected nodes and the remaining network nodes are enabled, and the damaged key nodes in the network are restored first. If the network is in a disconnected state, the nodes that are fastest connected to each connected subgraph are identified and used as target repair nodes; the target repair nodes are repaired first until the network forms a single connected subgraph so that the network reaches a connected state; when the network reaches a connected state, the target nodes that have the best network performance are selected in turn for restoration and the backup lines of the target nodes are enabled to obtain the optimal repair node sequence.

7. The power system network disaster simulation and restoration optimization method according to claim 6 is characterized in that: A hierarchical reinforcement learning model is constructed and used to perform node recovery, wherein the bottom layer network in the hierarchical reinforcement learning model preferentially repairs the target repair node until the network forms a single connected subgraph so that the network reaches a connected state, the state of the bottom layer network is defined as the number and size distribution of the remaining network connected subgraphs, and the action is defined as selecting a node for repair and enabling the backup line of the node to be repaired; the high-level network in the hierarchical reinforcement learning model is used to sequentially select the target node that optimizes the network capacity for recovery, the state of the high-level network is defined as the current remaining network and the set of nodes that need to be repaired and the connectivity measure of the entire network, and the action is defined as selecting a node that improves network performance for repair and enabling the backup line of the selected node.

8. The power system network disaster simulation and restoration optimization method according to claim 7 is characterized in that: The reward function of the underlying network is defined as , represents the maximum connected subgraph size of the network after the node is repaired, represents the number of connected subgraphs of the network after the node is repaired, and They represent the weight parameters for measuring the maximum size of the network's connected subgraph and the number of the network's connected subgraphs, respectively. Indicates the first repair node The average clustering coefficient of the posterior network, It is the weight to measure the priority of repairing key nodes. is the first indicator function, used when the first repair node Belongs to the priority repair key node set The value is 1 when it is, otherwise it is 0; The reward function of the high-level network is defined as , Indicates repair node The average clustering coefficient of the posterior network, Indicates that according to the second repair node The connectivity measure of the network is determined by the number of nodes in the largest connected subgraph in the remaining network except Indicates that except for the second repair node The remaining network, and represents the weight parameter, is the second indicator function, when the second repair node Belongs to the priority repair key node set The value is 1 when , otherwise it is 0.

9. The power system network disaster simulation and restoration optimization method according to claim 7 or 8, characterized in that: Using the hierarchical reinforcement learning model to perform node recovery includes: Building the underlying state table and high-level state table It is used to store the expected return value of each state-action pair of the underlying network and the high-level network; Obtaining a set of nodes removed from the power system network under different disaster types and remaining network data and inputting them into the hierarchical reinforcement learning model; In the hierarchical reinforcement learning model, the underlying network is in each state Next, according to the underlying state table and exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , Represents the discount factor; repeat the process until the number of connected subgraphs of the network becomes 1 after the node is repaired, and then switch to the high-level network for node recovery; When a high-level network performs node recovery, in each state According to the high-level status table And the exploration rate Select Action , with probability Select the underlying state table The action with the largest median value and the probability Randomly select an action and execute the selected action Then get the reward according to the reward function and the next state , and follow Update the underlying state table , repeat until the preset iteration stop condition is reached to determine the optimal underlying state table and high-level state table , and obtain a hierarchical reinforcement learning model with optimal parameters; The node set removed from the current power system network and the remaining network data are obtained and input into the hierarchical reinforcement learning model with optimal parameters to perform node recovery.

10. A computer device comprising a processor and a memory, wherein the memory is used to store a computer program, wherein: The processor is configured to execute the computer program to perform the method according to any one of claims 1 to 8.

Images