Weak link identification method and device based on power grid topological structure

Abstract

The invention discloses a weak link identification method and device based on a power grid topological structure, and the method comprises the steps: firstly building a node-side topological model of a power grid, enabling nodes to comprise a transformer, a section switch, a load point and a side expression line, and integrating electrical parameters and geographic information; and then, GNN is combined with CNN to perform feature extraction on the topological structure and the operation data, and a weakness score is generated, so that a weak link in the power grid is identified. And based on an identification result, generating a plurality of power grid improvement strategies by adopting a generative adversarial network (GAN), carrying out multi-objective optimization through deep reinforcement learning (DRL), and selecting an optimal improvement measure. The improvement measures are subjected to priority ranking in combination with factors such as implementation cost, reliability improvement and load balancing. The weak link of the power grid can be accurately identified, the efficient improvement strategy is automatically generated, and the overall reliability and operation efficiency of the power grid are improved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a method and device for identifying weak links based on the power grid topology structure, which are used to improve the power supply reliability and stability of the power grid. Background Art

[0002] With the development of the power system, the power grid structure has gradually become complex. The large-scale access of distributed power sources, the continuous increase in load, and the change in power consumption demand have brought challenges to the operation reliability of the distribution network. Traditional methods for identifying weak links in the power grid mainly focus on the analysis of single factors (such as power supply radius, load density), fail to comprehensively consider the comprehensive influence of multiple factors, and have a low accuracy in identifying weak links under complex topological structures. In addition, existing technologies usually have difficulty in automatically generating specific power grid improvement measures, thus limiting the rapid response ability of the system. Therefore, it is necessary to propose an improved method for identifying weak links based on the power grid topology structure to improve the reliability and response ability of the power grid.

[0003] For this reason, some attempts have been made in the prior art. For example, the prior art 1 with the publication number CN116523001A discloses a method and device for constructing a power grid weak line identification model, a computer device, a storage medium, and a computer program product. The method therein establishes power grid graph data based on historical power grid operation state data and historical topological information through a graph representation method; performs multi-head attention information aggregation analysis on the power grid graph data to obtain the multi-head attention information of the power grid graph data; and conducts key operation feature identification to determine the key operation features of the power grid operation; finally, through training an initial intelligent agent model, a power grid weak line identification model is obtained. During the model training process, a weighted cross-entropy function is used to form the loss function for training the graph attention network, and a weight correction coefficient is set for the lines with historical risk or fault events. However, there are problems such as a single model training process, lack of optimization solutions, and lack of global analysis ability in the prior art 1, which are specifically manifested as follows: 1. Using the multi-head attention mechanism and the graph attention network, although it can capture complex power grid graph information, the feature extraction process may be too dependent on historical data and preset weights, lacking the ability to dynamically adapt to the real-time changes of the power grid operation state, and insufficient support for the dynamic update and multi-objective optimization of the model; 2. The analysis of power grid weak points pays more attention to the risk assessment of single lines, and fails to fully combine the global topology and multi-factor assessment.

[0004] 3. It is only limited to the identification of weak lines, and no specific improvement strategies or optimization solutions are proposed, and it cannot directly guide the improvement of the power grid and the enhancement of reliability; Summary of the Invention To overcome the above problems existing in the prior art, the present invention provides a method and device for identifying weak links based on the power grid topology structure. Through multi-factor comprehensive evaluation, the accuracy of weak link identification is improved, and specific power grid optimization measures can be automatically generated to improve the power supply reliability and system stability of the power grid.

[0005] According to the first aspect of the present invention, there is provided a method for identifying weak links based on the power grid topology structure. The method includes the following steps: S1: Establish a power grid topology model including nodes and edges, where the nodes include transformers, sectional switches, and load points, and the edges represent lines. The power grid topology model contains the electrical parameters and geographical information of each node and edge; S2: Based on the power grid topology model, calculate the power supply radius of each load point and the weak link parameters of each line and sectional switch during the historical period. According to the calculation results, mark the load points and lines that do not meet the preset evaluation criteria as historical weak links; S3: Based on the topology model and historical weak links, establish and train a weak link scoring model based on a graph neural network and a deep learning algorithm to extract features from the real-time electrical parameters of the topology model to generate the real-time weak link scores of each node and edge; S4: Based on the real-time weak link scores of each node and edge, use a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism to dynamically identify the weak links in the power grid.

[0006] Further, the method further includes the following steps: S5: Utilize the historical power grid state data and the identified weak links, comprehensively consider power supply reliability, power supply cost, and load uniformity to design a reward function, train a GAN network to generate various power grid optimization strategies, and train a DRL agent to prioritize and select the optimal combination of power grid optimization strategies for the generated power grid optimization strategies; the various power grid optimization strategies include: adding sectional switches, optimizing the transformer layout, and adjusting the load transfer path; S6: According to the real-time power grid state, use the trained GAN network to generate various possible real-time power grid optimization strategies, and use the trained DRL agent to prioritize and select the optimal combination of real-time power grid optimization strategies for the real-time power grid optimization strategies.

[0007] Further, in step S2, the weak link parameters include the power supply radius of each load point and the number of transformers, load density, equipment failure rate, and circuit aging coefficient corresponding to each line and sectional switch; Among them, the power supply radius of each load point is determined by the following method: S21: Perform weighted averaging on all power supply paths from the main power supply node 𝑠 to the target load point 𝑖 to determine the uncorrected power supply radius , expressed as:

[0008] Among them, represents the length of the n th power supply path from the main power supply node 𝑠 to the target load point 𝑖, is the weight of each power supply path, N is the total number of power supply paths from the main power supply node 𝑠 to the target load point 𝑖; S22: Based on the load demands of each load point and the average load of all load points, correct the uncorrected power supply radius to obtain the power supply radius corresponding to the load point , expressed as:

[0009] Among them, is the power supply radius of the load point i , is the load demand of the load point, is the average load of all load points.

[0010] Furthermore, in step S2, according to the calculation results, the load points and lines that do not meet the preset evaluation criteria are marked as historical weak links, including: S21: Compare the power supply radius of each load point with the preset power supply radius threshold, and mark the load points with a power supply radius exceeding the power supply radius threshold as weak load points; S22: Perform a weighted sum of the number of transformers, load density, equipment failure rate, and line aging coefficient of each line to determine the comprehensive weak degree value of each line; compare the comprehensive weak degree value of each line with the preset comprehensive weak degree threshold, and mark the lines with a comprehensive weak degree value exceeding the comprehensive weak degree threshold as weak lines; S23: Take the weak load points and weak lines as weak links.

[0011] Furthermore, step S3 includes; S31: Use the power grid topology structure as the graph input of the graph neural network, and the node features including electrical parameters, load, failure rate, etc. as the input; S32: Use the adjacency matrix to represent the connection relationship between nodes, and use the graph neural network to update the feature representations of nodes and the feature representations of lines according to the adjacency matrix through multiple graph convolutional layers; S33: Use a convolutional neural network CNN or a fully connected neural network to further extract and process the updated feature representations of nodes and lines, generate the weak score representations of each node and line, and form the weak score model; S34: Train the vulnerability scoring model using supervised learning based on the topological model and historical weak links.

[0012] Further, step S4 includes: S41: Calculate the real-time vulnerability scores of each node or edge in the power grid using the vulnerability scoring model; S42: Define the fitness function of the genetic algorithm combined with local search using the vulnerability scores and the vulnerability characteristics of each node and line to evaluate whether a node or edge belongs to a weak link, and obtain the initial optimal static weak link set; S43: Conduct an importance analysis of the vulnerability characteristics through the random forest algorithm, and adjust the weight parameters of the fitness function in the genetic algorithm combined with local search according to the analysis results, so that the higher the importance of the vulnerability characteristic, the greater the weight parameter; S44: Update the optimal static weak link set using the genetic algorithm combined with local search according to the adjusted fitness function; S45: Iteratively execute steps S43 and S44 to obtain the final weak link set; S46: Construct time series features based on the real-time vulnerability scores and each vulnerability characteristic, calculate the vulnerability probabilities of each weak link in the optimal weak link set through the random forest algorithm introducing a voting mechanism based on the time series features, and mark the node or line with the highest vulnerability probability as the final weak link.

[0013] Further, in step S42, the genetic algorithm combined with local search includes: Initialize the population, generate a number of random solutions, and each solution corresponds to a possible weak link set; Evaluate the quality of each solution using the improved fitness function, and then select the optimal solution according to the fitness function and perform crossover and mutation; Introduce the local search method to locally optimize the solutions of each generation until the fitness function converges and find the optimal static weak link set.

[0014] Further, in step S43, the importance analysis of the vulnerability characteristics through the random forest algorithm includes: Determine the importance of the j th feature in the random forest according to the following formula :

[0015] where is the information gain brought by the feature k in the th decision tree,K is the total number of decision trees.

[0016] Furthermore, in step S46, the vulnerability probability of each weak link in the optimal weak link set is calculated by the following formula:

[0017] where is the vulnerability probability of node i; is the true label of node i , = 1 indicates that node i is a weak link, = 0 indicates that node i is normal; is the probability that node i is a weak link given by the k th decision tree; is the weight of the k th decision tree, K is the total number of decision trees.

[0018] Furthermore, in step S5, a reward function is designed by comprehensively considering power supply reliability, power supply cost, and load uniformity, including: S51: Taking the sum of the power supply radii of all nodes in the power grid after adding sectional switches to be the smallest as the goal, a first optimization objective function is obtained, where the positions of the added sectional switches are determined using the SA-PSO algorithm based on the identified weak links; the first optimization objective function is expressed as:

[0019] where is the power supply radius of node i in the power grid; x is the position of the added sectional switch; is the cost of adding sectional switches; is the weight for balancing the power supply radius and the switch cost; S52: According to the load distribution, a second optimization objective function is obtained with the goal of maximizing the balance of the loads of all transformers, expressed as:

[0020] where is the load of transformer m ; is the total number of transformers; is the layout position of the transformer; S53: Obtain the third optimization objective function with the goal of the most balanced load distribution after the load transfer mechanism is started; the load transfer mechanism is started when the load on any line or node exceeds the preset threshold; the third optimization objective function is expressed as:

[0021] where, represents the load of the line after load transfer n ; is the average load of all lines; is the selection variable of the load transfer path; S54: Perform weighted summation on the first to third optimization objective functions to obtain the reward function.

[0022] According to the second aspect of the present invention, there is provided a weak link identification device based on the power grid topology structure. This device adopts the weak link identification method based on the power grid topology structure as described in the first aspect of the present invention. The device includes: A topology model establishment module, used to establish a power grid topology model including nodes and edges, where the nodes include transformers, sectional switches, and load points, and the edges represent lines. The power grid topology model contains the electrical parameters and geographical information of each node and edge; A weak link marking module, used to calculate the weak parameter corresponding to each load point, each line, and sectional switch within the historical period based on the power grid topology model, and mark the load points, lines, and sectional switches that do not meet the preset evaluation criteria as historical weak links according to the calculation results; A scoring model establishment module, used to establish and train a weak link scoring model based on graph neural network and deep learning based on the topology model and historical weak links, and extract features from the real-time electrical parameters of the topology model to generate the real-time weak link scores of each node and edge; An identification module, used to dynamically identify the weak links in the power grid based on the real-time weak link scores of each node and edge, adopting a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism.

[0023] Furthermore, the device further includes: A training module, used to utilize the historical power grid state data and the identified weak links, comprehensively consider power supply reliability, power supply cost, and load uniformity to design a reward function, train a GAN network to generate various power grid optimization strategies, and train a DRL agent to perform priority ranking on the generated power grid optimization strategies and select the optimal combination of power grid optimization strategies; the various power grid optimization strategies include: adding sectional switches, optimizing transformer layout, and adjusting load transfer paths; An optimization module, configured to generate various possible real-time power grid optimization strategies by using a trained GAN network according to the real-time power grid state, and use a trained DRL agent to rank the real-time power grid optimization strategies and select an optimal combination of real-time power grid optimization strategies.

[0024] According to a third aspect of the present invention, there is provided a terminal. The terminal includes a processor and a storage medium; The storage medium is used for storing instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to the first aspect of the present invention.

[0025] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps of the method according to the first aspect of the present invention are implemented.

[0026] Compared with the prior art, the present invention has the following beneficial effects: Based on a topological model and historical weak links, a vulnerability scoring model based on a graph neural network and a deep learning algorithm is established and trained, which is used to extract features from the real-time electrical parameters of the topological model to generate a vulnerability score; a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism are used to determine the real-time vulnerability scores of each node and edge of the power grid, so as to dynamically identify the weak links in the power grid. In this way, the weak links in the power grid can be identified more accurately; the dynamic weight adjustment mechanism improves the accuracy of the vulnerability score. In addition, the present invention can generate various possible real-time power grid optimization strategies by using a trained GAN network according to the real-time power grid state, and use a trained DRL agent to rank the real-time power grid optimization strategies and select an optimal combination of real-time power grid optimization strategies. In this way, the manual intervention is reduced by the automatically generated power grid optimization measures, and the rapid response ability of the power grid is improved; the evolution trend analysis and risk warning method provide forward-looking protection measures for the power grid, further enhancing the power supply reliability and stability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic flowchart of a method for identifying weak links based on a power grid topological structure in an embodiment of the present invention; Figure 2 It is a more specific schematic flowchart of a method for identifying weak links of a power grid based on a power grid topological structure of the present invention; Figure 3 It is a schematic diagram of a power grid topological structure, including the identifications of lines, transformers, sectional switches and load nodes; Figures 4(a) and 4(b) are respectively the analysis of the economic cost of planning configuration under rigid and elastic reliability requirements; Figure 5 Schematic diagram of an automatically generated power grid planning scheme based on weak link identification results. Specific implementation manners

[0028] The present invention will be further illustrated below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent modifications made by those skilled in the art fall within the scope defined by the appended claims of this application.

[0029] In a first aspect of the present invention, a method for identifying weak links based on a power grid topological structure is provided.

[0030] As Figure 1 , in one embodiment, the method includes the following steps: S1: Establish a power grid topological model including nodes and edges, where the nodes include transformers, sectional switches, and load points, and the edges represent lines. The power grid topological model contains the electrical parameters and geographical information of each node and edge.

[0031] S2: Based on the power grid topological model, calculate the weak link parameters corresponding to each load point, each line, and sectional switch during the historical period, and mark the load points, lines, and sectional switches that do not meet the preset evaluation criteria as historical weak links according to the calculation results.

[0032] Among them, the weak link parameters may include the power supply radius of each load point, and the number of transformers, load density, equipment failure rate, and circuit aging coefficient corresponding to each line and sectional switch.

[0033] The calculation methods of each weak link parameter are introduced separately below.

[0034] A. Power supply radius The power supply radius of each load point refers to the shortest path distance from the substation (or main power source node) to the load point. To calculate the power supply radius, the shortest path Dijkstra algorithm is used.

[0035] First, calculate the shortest path distance from the main power source node 𝑠 to the target load point 𝑖. The calculation formula is:

[0036] Among them, represents the actual length of each path from the main power source node s to the target load point i through the lines (i.e., through the edges), the calculated shortest path, path represents each path from the main power source node s to the target load point i , and N represents the set of power source nodes.

[0037] If there are multiple alternative paths for a certain load point i the case of the backup path is considered.

[0038] Therefore, the power supply radius of each load point is determined as follows: 1) The uncorrected power supply radius is determined by weighted averaging all power supply paths from the main power supply node 𝑠 to the target load point 𝑖 , expressed as:

[0039] where represents the length of the n th power supply path from the main power supply node 𝑠 to the target load point 𝑖, is the weight of each power supply path, N is the total number of paths from the main power supply node 𝑠 to the target load point 𝑖. Among them, the weight can be obtained based on the reliability of the path or the load conditions.

[0040] 2) The uncorrected power supply radius is corrected according to the load demand of each load point and the average load of all load points to obtain the power supply radius corresponding to the load point, expressed as:

[0041] [[ID=3�]]where is the load demand of the load point, is the average load of all load points. Load points with larger loads will have a larger correction factor to reflect their higher power supply requirements.

[0042] B. Number of transformers The number of distribution transformers connected to each line and the sectional switches included can be directly counted. Too many distribution transformers will increase the load pressure on the line or sectional switch and affect the reliability. First, identify the nodes connected to each line or sectional switch and distinguish the node types. Distribution transformers are a type of grid node, so it is necessary to identify which nodes are distribution transformers. For each line or sectional switch, the number of distribution transformers connected to the line can be counted by statistically analyzing all its connected nodes. The specific calculation formula is as follows:

[0043] where L i represents the set of all nodes connected to the line; T j represents the type of node j when Tj When it is = 1, it means that this node is a distribution transformer.

[0044] C. Load density The load density on each line or sectionalizing switch is expressed as the connected load and the length of this line ratio:

[0045] Among them, represents the total load connected to the line (including the sum of the loads of all nodes), is the actual length of this line. Lines with higher load density are more likely to have overload problems, affecting reliability.

[0046] D. Equipment failure rate The equipment failure rate can be defined by the ratio of the number of failures of the line or sectionalizing switch in a recent period of time to the total operating time:

[0047] The higher the failure rate, the worse the stability and reliability of this equipment or line.

[0048] E. Equipment failure rate The degree of line aging is represented by the actual service life of the line and its designed service life and the aging coefficient calculation formula is:

[0049] When approaches or is greater than 1, it indicates that the line is severely aged and renewal or maintenance needs to be considered.

[0050] In this step S2, according to the calculation results, the load points and lines that do not meet the preset evaluation criteria are marked as historical weak links, which specifically include the following steps: S21: Compare the power supply radius of each load point with the preset power supply radius threshold, and mark the load points whose power supply radius exceeds the power supply radius threshold as weak load points.

[0051] Specifically, this step includes setting an upper threshold for the power supply radius , when the power supply radius of a certain load point exceeds this threshold (that is, when ), mark this node as a weak load point with an overly long power supply radius.

[0052] S22: Perform a weighted sum of the number of transformers, load density, equipment failure rate, and line aging coefficient for each line to determine the comprehensive weakness degree value of each line; compare the comprehensive weakness degree value of each line with a preset comprehensive weakness threshold value, and mark the lines whose comprehensive weakness degree value exceeds the comprehensive weakness threshold value as weak lines.

[0053] Among them, the comprehensive weakness score of the line i is calculated by weighting , and can be achieved through the following expression:

[0054] Among them, represents the number of distribution transformers, is the load density, is the equipment failure rate, is the line aging coefficient, , , , are the weight coefficients respectively, and are adjusted according to the actual power grid situation to reflect the influence degree of each factor on the weakness score.

[0055] Set a threshold for the weakness score . When the comprehensive weakness score of a certain line or sectional switch exceeds this threshold, mark this line as a weak line:

[0056] S23: Take the weak load points and weak lines as historical weak links.

[0057] S3: Based on the topological model and historical weak links, establish and train a weakness scoring model based on graph neural networks (Graph Neural Networks, GNN) and deep learning algorithms, which is used to extract features from the real-time electrical parameters of the topological model to generate the real-time weakness scores of each node and edge.

[0058] The above steps mark some weak links. Next, by combining graph neural networks (Graph Neural Networks, GNN) and deep learning methods, more features can be further extracted from the topological structure of the power grid and its operation data, and the identification of weak links can be optimized and refined. The advantage of GNN is that it can capture the mutual influence between nodes and their neighbor nodes through adjacency relationships, so it can analyze the overall operation of the power grid more comprehensively.

[0059] Specifically, step S3 may include the following steps: S31: Use the power grid topology structure as the graph input of the graph neural network, and the node features include electrical parameters, load, and failure rate as inputs.

[0060] Among them, the electrical parameters may include, for example, voltage, current, etc.

[0061] S32: Use the adjacency matrix to represent the connection relationship between nodes, and use the graph neural network to update the feature representations of nodes and the feature representations of lines through multiple graph convolution layers according to the adjacency matrix.

[0062] In this step, the feature representations of nodes and lines are respectively the feature vectors of nodes and the feature vectors of lines. Through the graph convolution layer, high-level features of each node and line can be extracted, and these high-level features consider the attributes and connection relationships of nodes and their neighbor nodes.

[0063] Specifically, the adjacency matrix can be represented as A, and the elements in the matrix take the following values:

[0064] Through the graph convolution operation of multiple graph convolution layers, the feature vector of each node is updated, and high-level features considering the attributes of neighbor nodes can be extracted. The feature of node i at the l layer is updated as:

[0065] where, is the feature representation of node l at the i layer, is the feature representation of node l at the j layer, represents the set of neighbor nodes of node i , is the weight matrix of the graph convolution layer at the l layer, is the bias vector at the l layer, is the activation function.

[0066] Through the graph convolution operation of multiple graph convolution layers, the feature vector of each line is updated, and high-level features considering the connection relationships of neighbor nodes can be extracted. The update of the feature representation of line i to j is based on the features of the two connected nodes (source node and target node) and the feature of the line itself, and its update formula is as follows:

[0067] Among them, is the feature vector of the l layer of circuits i to j . This feature vector is the result of dynamic calculation, which will be updated as the network propagates. It is used for weak link identification and can be updated iteratively through the neural network algorithm; is the feature vector of the l source node of the i -1 layer; is the feature vector of the l target node of the j -1 layer; is the feature vector of the l circuit of the i to j itself (including parameters such as length, impedance, capacity, and overload factor that will not be directly optimized or modified by the neural network algorithm); is the weight matrix of the l layer; is the bias vector of the l layer; (·) is the activation function, and the sigmoid function is used in this patent.

[0068] Table I shows the node feature update method and the circuit feature update method to better illustrate the difference between the two.

[0069] Table I

[0070] S33: Use a convolutional neural network CNN or a fully connected neural network to perform further feature extraction and processing on the updated node feature representation and circuit feature representation, generate the vulnerability score representation of each node and circuit, and form the vulnerability score model.

[0071] Specifically, taking the fully connected neural network (MLP) as an example, the updated node feature representation and the circuit feature representation can be input into the deep learning model, and the fully connected neural network is used to perform further feature processing and classification. The specific operations include: Input the updated node feature representation extracted by the GNN and the circuit feature representation

[0072] [[ID=..]]

[0073] The vulnerability scores of each node can be generated and the vulnerability scores of each line , and it is determined whether they belong to weak links according to the topological structure and characteristics of the nodes and lines.

[0074] S34: Based on the topological model and historical weak links, use supervised learning to train the vulnerability scoring model.

[0075] Specifically, this step includes: Step S341: Use the cross-entropy loss function to measure the difference between the weak links predicted by the model and the true labels determined by the historical weak links; where the cross-entropy loss function is defined as:

[0076] where, is the predicted vulnerability score of node i, obtained after passing through the Sigmoid activation; is the true label; N is the total number of links corresponding to all nodes and lines.

[0077] Preferably, it can be set that when the true label = 1, it indicates a weak link, and when the true label = 0, it indicates a normal link.

[0078] Step S342: Minimize the cross-entropy loss function through the Adam optimization algorithm and adjust the parameters in the vulnerability scoring model.

[0079] In this way, after training, the model can generate the vulnerability scores of each node and line according to the real-time data or historical data of the power grid.

[0080] S4: Based on the real-time vulnerability scores of each node and line, adopt an improved genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism to dynamically identify the weak links in the power grid.

[0081] In this step, the purpose of the improved genetic algorithm combined with local search is to evaluate the probability of each node or line being a weak link by obtaining the result of weighted voting through the fitness function, and mark the node or line with the highest weak link probability as the weak link. The purpose of the random forest algorithm with a voting mechanism is to further analyze the influence of each input feature (such as power supply radius, load density, equipment failure rate, etc.) on the weakness of nodes or edges. In particular, the importance of each input feature is determined by calculating the average information gain of each feature in all trees, which helps to screen out the features that have the greatest impact on the prediction result and simplify the model. The result of feature importance analysis by the random forest can also be used to verify and improve the result of the genetic algorithm.

[0082] Specifically, this step S4 includes: S41: Calculate the real-time weakness score of each node or edge in the power grid using the weakness scoring model.

[0083] Among them, the weakness scoring model here uses the weakness scoring model trained in step S3.

[0084] S42: Define the fitness function of the genetic algorithm combined with local search using the weakness score and the weakness characteristics of each node and line to evaluate whether a node or edge belongs to a weak link, and obtain the initial optimal static weak link set.

[0085] As an example, this fitness function can be constructed as:

[0086] Among them, x represents a set of candidate weak links, S i is the comprehensive weakness score of node i (or line i), is the power supply radius of node i, normalized to , that is, divided by the maximum power supply radius to make its value between 0 and 1; is the number of transformers on line i, normalized to ; is the load density, normalized to ; is the equipment failure rate, normalized to ; is the line aging coefficient, normalized to , and normalization processing is adopted to ensure the stability and universality of the optimization calculation; , , , , , represent the corresponding weight parameters respectively.

[0087] This fitness function can comprehensively evaluate the advantages and disadvantages of the set of candidate weak links.

[0088] Specifically, the genetic algorithm combined with local search may include: initializing the population, generating a number of random solutions, with each solution corresponding to a possible set of weak links; using the improved fitness function to evaluate the advantages and disadvantages of each solution, and then selecting the optimal solution according to the fitness function and performing crossover and mutation; introducing a local search method to locally optimize the solutions of each generation to further improve the quality of the solutions until the fitness function converges to find the optimal set of static weak links.

[0089] S43: Perform importance analysis on the vulnerability features through the random forest algorithm, and adjust the weight parameters of the fitness function in the genetic algorithm combined with local search according to the analysis results, so that the higher the importance of the vulnerability feature, the greater the weight parameter.

[0090] In the identification of weak links in the power grid, different features have different importance for vulnerability judgment. Through feature selection and importance analysis, the features that have the greatest impact on the model prediction results can be screened out, and the model can be simplified.

[0091] Suppose the feature is the j th feature, and the feature importance in the random forest can be determined by calculating the average information gain of this feature in all trees:

[0092] where is the information gain brought by the feature k in the th decision tree, and K is the total number of decision trees. By calculating , the weight parameters corresponding to each vulnerability feature of the fitness function in the genetic algorithm combined with local search can be adjusted, so that the higher the importance of the feature, the greater the weight parameter.

[0093] S44: According to the adjusted fitness function, use the genetic algorithm combined with local search to update the optimal set of static weak links.

[0094] This step is the same as the method for updating weak links in step S42 and will not be elaborated here.

[0095] S45: Iteratively execute steps S43 and S44 to obtain the final set of static weak links.

[0096] Among them, the iteration can be stopped when the result converges.

[0097] S46: Construct time series features based on the real-time vulnerability scores and various vulnerability characteristics. Calculate the vulnerability probabilities of each vulnerable link in the optimal set of vulnerable links based on the time series features through a random forest algorithm with a voting mechanism, and mark the node or line with the highest vulnerability probability as the final vulnerable link.

[0098] In this step, by constructing time series features, the dynamic changes of the weak links in the power grid can be captured, and the prediction ability of the model for time-correlated data can be improved. Among them, the vulnerability probabilities of each vulnerable link in the optimal set of vulnerable links are calculated by the following formula:

[0099] where, is the vulnerability probability of node i; is the probability that node i is a vulnerable link given by the k-th decision tree; is the weight of the k-th decision tree, usually related to its accuracy or feature importance; K is the total number of decision trees.

[0100] Based on the result of weighted voting, mark the node or line with the highest vulnerability probability as the vulnerable link.

[0101] S5: Utilize the historical power grid state data and the identified vulnerable links, comprehensively consider power supply reliability, power supply cost, and load uniformity to design a reward function, train a GAN (Generative Adversarial Network) to generate various power grid optimization strategies, and train a DRL agent to prioritize the generated power grid optimization strategies and select the optimal combination of power grid optimization strategies; the various power grid optimization strategies include: adding sectional switches, optimizing transformer layout, and adjusting load transfer paths.

[0102] By generating strategies through the GAN network and then optimizing the strategies by the DRL agent, the collaborative work of the generation model and the decision-making model is realized, which is a multi-modal learning method. Its steps include: Design the generator of the GAN network to generate power grid improvement strategies, including reconfiguration of transformers, layout of new lines, load distribution, etc. Design the discriminator of the GAN network to evaluate the effectiveness of the generated strategies. Train based on the historical data and current state of the power grid to distinguish good strategies from bad strategies.

[0103] Define the state of the power grid as the state of the DRL environment, including the current configuration of the power grid, load demand, power supply radius, etc. The power grid improvement strategy can be used as an action in the DRL network. Design a reward function to reflect the effectiveness of the power grid improvement strategy. The reward function can be designed based on multiple objectives such as power supply reliability, cost, and load balancing. Use the DRL network to train an agent so that it can learn to select the best strategy in the power grid environment. The decision-making network of the agent can be a deep neural network, which can process high-dimensional power grid state inputs and output the optimal action (i.e., the power grid improvement strategy).

[0104] The generator continuously generates new power grid improvement strategies. The DRL agent evaluates these strategies and selects the best strategy for implementation. The decision of the agent is fed back to the generator, and the generator improves its strategy generation ability based on these feedbacks.

[0105] Among them, the design of the reward function can include the following steps: S51: Obtain the first optimization objective function with the goal of minimizing the sum of the power supply radii of each node in the power grid after adding sectional switches. The positions of the added sectional switches are determined using the SA-PSO algorithm based on the identified weak links.

[0106] Determine the positions of the added sectional switches to reduce the power supply radius in certain areas, balance the load distribution, and reduce the impact of faults. Specifically, the first optimization objective function can be expressed as:

[0107] Among them, is the power supply radius of each node in the power grid i ; is the cost of adding sectional switches, x is the position of the added sectional switch, is the weight for balancing the power supply radius and the switch cost. By optimizing the model and choosing to add sectional switches in areas with a larger power supply radius, the power grid structure can be optimized.

[0108] S52: According to the load distribution situation, obtain the second optimization objective function with the goal of maximizing the balance of the loads of each transformer.

[0109] By setting the optimization objective to maximize the balance of the loads of each transformer, the overload situation can be reduced. Among them, the second optimization objective function can be expressed as:

[0110] Among them, is the load of transformer m ; is the total number of transformers; It is the layout position of the transformer. By optimizing the algorithm to adjust the layout of the transformer, it is ensured that the loads of each transformer are more balanced and the overload problem is avoided.

[0111] S53: Obtain the third optimization objective function with the goal of the most balanced load distribution after the load transfer mechanism is started; the load transfer mechanism is started when the load on any line or node exceeds the preset threshold.

[0112] Optimizing the load transfer path can make the load distribution after transfer more balanced and avoid local line or equipment overload. Specifically, the third optimization objective function is:

[0113] where, represents the load of the line after load transfer n ; is the average load of all lines; is the selection variable of the load transfer path, and the local overload risk is reduced by dynamically adjusting the load through transfer.

[0114] S54: Perform weighted summation on the first to third optimization objective functions to obtain the reward function.

[0115] Combining the above three optimization measures, the system can automatically generate an optimization strategy for the power grid according to the real-time data and the results of weak link identification.

[0116] The DRL agent can also prioritize the generated improvement strategies. Specifically, a priority score can be assigned to each generated power grid improvement strategy according to factors such as implementation cost, reliability improvement, and load balance; all strategies are sorted according to the priority scores of each strategy to determine the order of implementation.

[0117] The location and implementation order of the optimization measures can be displayed by providing a visual interface.

[0118] S6: According to the real-time power grid state, use the trained GAN network to generate multiple possible real-time power grid optimization strategies, and use the trained DRL agent to prioritize and select the optimal combination of real-time power grid optimization strategies for the real-time power grid optimization strategies.

[0119] A more specific flow diagram of a method for identifying weak links in a power grid based on the power grid topology structure of the present invention is as Figure 2 shown.

[0120] Based on the above scheme, in order to verify the effectiveness of the method of the present invention, in this embodiment, it is verified through experiments as follows: The historical data of the topology structure of an actual 10 kV line in Lianyungang City, Jiangsu Province was used in the experimental part. Among them, the historical collected data was from 0:00 on June 14, 2024 to 0:00 on June 16, 2024, and the sampling frequency was 15 min / point.

[0121] First, the power grid topology structure file was parsed using the Matlab simulation platform, and key equipment such as lines, transformers, loads, and sectional switches were marked as Figure 3 shown. Figure 3 It overall shows the equipment distribution and load aggregation under this actual line.

[0122] Next, by the above method, information such as the number of distribution transformers, load density, line aging degree, and the number of sectional switches in a single line was counted to evaluate the power supply reliability of the line. For lines with insufficient power supply reliability, the rigid and elastic reliability requirements of user nodes were considered separately, and the types of equipment for joint optimal configuration gradually increased. In the case of rigid reliability requirements, when the types of equipment for joint optimal configuration were few, the flexibility of reliability service supply was insufficient, and the rigid reliability requirements of some users could not be met; when the types of equipment for joint optimal configuration were sufficient, the rigid reliability requirements of all users were met, indicating that the proposed unified optimal configuration method for multiple types of equipment can make full use of the characteristics of different equipment and flexibly meet the rigid reliability requirements of users. In addition, as shown in Fig. 4(a), as the types of equipment increase, the total investment operation and maintenance cost of equipment to meet the same rigid reliability requirement decreases, reflecting that the method proposed in the present invention can give full play to the combined advantages of different equipment and effectively improve the economy of equipment investment planning. As shown in Fig. 4(b), in the case of elastic reliability requirements, as the types of equipment increase, the total system cost considering the total investment operation and maintenance cost of equipment and the user power outage cost decreases significantly. Compared with the case where the only optional equipment is a manual sectional switch, adding an automatic sectional switch, a manual external tie line, an automatic external tie line, a manual internal tie line, an automatic internal tie line, a circuit breaker, a spare line, and a load control switch in turn, the total system cost decreases by 11.6%, 36.4%, 45.1%, 50.7%, 52.8%, 55.2%, 67.1%, and 69.1% respectively, indicating that the proposed method can effectively improve the economy of equipment configuration and the flexibility of reliability service supply. [[ID=1,2]]

[0123] According to the evaluation results of different reliability requirements and the weakness of the power supply line, the result diagram of the comprehensive optimal configuration considering reliability and economy in a local area of the 10 kV actual line is as Figure 5 shown. It shows that the proposed unified optimal configuration method for multiple types of equipment can give full play to the characteristics and combined advantages of various equipment, flexibly, economically, and efficiently meet the differentiated reliability requirements of users, fully improve the equipment utilization efficiency, and effectively reduce the total system cost.

[0124] In the second aspect of the present invention, an improved device for identifying weak links based on the power grid topology structure is provided. The method for identifying weak links based on the power grid topology structure as described in the first aspect of the present invention is adopted. The device includes: A topology model establishment module, which is used to establish a power grid topology model including nodes and edges, where the nodes include transformers, sectional switches, and load points, and the edges represent lines. The power grid topology model contains the electrical parameters and geographical information of each node and edge; A weak link marking module, which is used to calculate the weak parameter corresponding to each load point, each line, and sectional switch within the historical period based on the power grid topology model, and mark the load points, lines, and sectional switches that do not meet the preset evaluation criteria as historical weak links according to the calculation results; A scoring model establishment module, which is used to establish and train a weak link scoring model based on graph neural network and deep learning based on the topology model and historical weak links, and is used to extract features from the real-time electrical parameters of the topology model to generate the real-time weak link scores of each node and edge; An identification module, which is used to dynamically identify the weak links in the power grid based on the real-time weak link scores of each node and edge, and adopt a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism.

[0125] Further, the device further includes: A training module, which is used to utilize the historical power grid state data and the identified weak links, comprehensively consider the power supply reliability, power supply cost, and load uniformity to design a reward function, train a GAN network to generate various power grid optimization strategies, and train a DRL agent to prioritize and select the optimal combination of power grid optimization strategies for the generated power grid optimization strategies; the various power grid optimization strategies include: adding sectional switches, optimizing the transformer layout, and adjusting the load transfer path; An optimization module, which is used to generate various possible real-time power grid optimization strategies by using the trained GAN network according to the real-time power grid state, and prioritize and select the optimal combination of real-time power grid optimization strategies by using the trained DRL agent.

[0126] According to the third aspect of the present invention, a terminal is provided. The terminal includes a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method as described in the first aspect of the present invention. The specific steps are not elaborated here.

[0127] According to the fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of the method as described in the first aspect of the present invention are implemented. The specific steps are not elaborated here.

[0128] The beneficial effects of the present invention are as follows. Compared with the prior art, based on the topological model and historical weak links, a vulnerability scoring model based on graph neural network and deep learning algorithm is established and trained to extract features from the real-time electrical parameters of the topological model to generate vulnerability scores; the genetic algorithm combined with local search and the random forest algorithm introducing a voting mechanism are used to determine the real-time vulnerability scores of each node and edge in the power grid, so as to dynamically identify the weak links in the power grid. In this way, the weak links in the power grid can be identified more accurately; the dynamic weight adjustment mechanism improves the accuracy of vulnerability scoring. In addition, the present invention can generate various possible real-time power grid optimization strategies by using the trained GAN network according to the real-time power grid state, and use the trained DRL agent to prioritize the real-time power grid optimization strategies and select the optimal combination of real-time power grid optimization strategies. In this way, manual intervention is reduced through the automatically generated power grid optimization measures, and the rapid response ability of the power grid is improved; the evolution trend analysis and risk warning methods provide forward-looking protection measures for the power grid, further enhancing the power supply reliability and stability of the power grid. [[ID=III]] [[ID=IV]]

[0129] The present disclosure may be a system, method, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to implement various aspects of the present disclosure. [[ID=VI]] [[ID=VII]]

[0130] A computer-readable storage medium may be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device such as a punch card or raised structures in a groove having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium used herein is not construed as an instantaneous signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire. [[ID=IX]] [[ID=X]]

[0131] The computer-readable program instructions described herein can be downloaded to various computing / processing devices from a computer-readable storage medium or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0132] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state-setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of the computer-readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer-readable program instructions to implement various aspects of the present disclosure.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the specific implementation manners of the present invention can still be modified or equivalently replaced, and any modification or equivalent replacement without departing from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for identifying weak links based on the power grid topological structure, characterized in that It includes the following steps: S1: Establish a power grid topology model including nodes and edges, where the nodes include transformers, sectionalizing switches, and load points, and the edges represent lines. The power grid topology model contains the electrical parameters and geographical information of each node and edge; S2: Based on the power grid topology model, calculate the vulnerability parameters corresponding to each load point, each line, and sectionalizing switch during the historical period. According to the calculation results, mark the load points, lines, and sectionalizing switches that do not meet the preset evaluation criteria as historical vulnerable links; S3: Based on the topology model and historical vulnerable links, establish and train a vulnerability scoring model based on graph neural network and deep learning algorithm to extract features from the real-time electrical parameters of the topology model to generate real-time vulnerability scores for each node and edge; S4: Based on the real-time vulnerability scores of each node and edge, use a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism to dynamically identify vulnerable links in the power grid.

2. The improved method for identifying weak links based on the power grid topology structure according to claim 1, wherein It also includes the following steps: S5: Utilize the historical power grid state data and the identified vulnerable links, comprehensively consider power supply reliability, power supply cost, and load uniformity to design a reward function, train a GAN network to generate multiple power grid optimization strategies, and train a DRL agent to prioritize and select the optimal combination of power grid optimization strategies for the generated power grid optimization strategies; The multiple power grid optimization strategies include: adding sectionalizing switches, optimizing transformer layout, and adjusting load transfer paths; S6: According to the real-time power grid state, use the trained GAN network to generate multiple possible real-time power grid optimization strategies, and use the trained DRL agent to prioritize and select the optimal combination of real-time power grid optimization strategies for the real-time power grid optimization strategies.

3. The method for identifying vulnerable links based on the power grid topology structure according to claim 1, wherein in step S2, the vulnerability parameters include the power supply radius of each load point and the number of transformers, load density, equipment failure rate, and circuit aging coefficient corresponding to each line and sectionalizing switch; wherein, the power supply radius of each load point is determined by the following method: S21: Determine the uncorrected power supply radius by weighted averaging all power supply paths from the main power supply node 𝑠 to the target load point 𝑖, which is expressed as: , expressed as: Among them, represents the length of the n th power supply path from the main power supply node 𝑠 to the target load point 𝑖, is the weight of each power supply path, N is the total number of power supply paths from the main power supply node 𝑠 to the target load point 𝑖; S22: Based on the load demands of each load point and the average load of all load points, correct the uncorrected power supply radius to obtain the power supply radius corresponding to the load point , which is expressed as: Among them, is the load demand of the load point, is the average load of all load points.

4. The method for identifying vulnerable links based on the power grid topology structure according to claim 2, wherein in step S2, marking the load points and lines that do not meet the preset evaluation criteria as historical vulnerable links according to the calculation results includes: S21: Compare the power supply radius of each load point with the preset power supply radius threshold, and mark the load points with a power supply radius exceeding the power supply radius threshold as vulnerable load points; S22: Perform weighted summation on the number of transformers, load density, equipment failure rate, and line aging coefficient of each line to determine the comprehensive vulnerability degree value of each line; compare the comprehensive vulnerability degree value of each line with the preset comprehensive vulnerability degree threshold, and mark the lines with a comprehensive vulnerability degree value exceeding the comprehensive vulnerability degree threshold as vulnerable lines; S23: Take the vulnerable load points and vulnerable lines as historical vulnerable links.

5. The method for identifying vulnerable links based on the power grid topology structure according to claim 1, wherein: step S3 includes; S31: Take the power grid topology structure as the graph input of the graph neural network, and the node features include electrical parameters, load, and failure rate as inputs; S32: Use the incidence matrix to represent the connection relationship between nodes, and use the graph neural network to update the feature representations of nodes and the feature representations of lines through multiple graph convolutional layers according to the incidence matrix; S33: Use a convolutional neural network CNN or a fully connected neural network to perform further feature extraction and processing on the updated feature representations of nodes and lines, generate the vulnerability score representations of each node and line, and form the vulnerability score model; S34: Based on the topology model and historical weak links, use supervised learning to train the vulnerability score model.

6. The method for identifying weak links based on the power grid topology structure according to claim 1, wherein Step S4 includes: S41: Calculate the real-time vulnerability scores of each node or edge in the power grid using the vulnerability score model; S42: Define the fitness function of the genetic algorithm combined with local search using the vulnerability scores and the vulnerability characteristics of each node and line to evaluate whether a node or edge belongs to a weak link, and obtain the initial optimal static weak link set; S43: Perform importance analysis on the vulnerability characteristics through the random forest algorithm, and adjust the weight parameters of the fitness function in the genetic algorithm combined with local search according to the analysis results, so that the vulnerability characteristics with higher importance have larger weight parameters; S44: Update the optimal static weak link set using the genetic algorithm combined with local search according to the adjusted fitness function; S45: Iteratively execute steps S43 and S44 to obtain the final static weak link set; S46: Construct time series features based on the real-time vulnerability scores and each vulnerability characteristic, calculate the vulnerability probabilities of each weak link in the optimal weak link set through the random forest algorithm introducing a voting mechanism based on the time series features, and mark the node or line with the highest vulnerability probability as the final weak link.

7. The method for identifying weak links based on the power grid topology structure according to claim 6, wherein In step S42, the genetic algorithm combined with local search includes: Initialize the population, generate several random solutions, and each solution corresponds to a possible weak link set; Evaluate the quality of each solution using the improved fitness function, and then select the optimal solution according to the fitness function and perform crossover and mutation; Introduce a local search method to locally optimize the solutions of each generation until the fitness function converges to find the optimal static weak link set.

8. The method for identifying weak links based on the power grid topology structure according to claim 6, wherein In step S43, performing importance analysis on the vulnerability characteristics through the random forest algorithm includes: Determine the feature importance of the j th feature in the random forest according to the following formula : Among them, is the k information gain brought by feature in the K th decision tree, and K is the total number of decision trees.

9. The method for identifying weak links based on the power grid topology structure according to claim 6, wherein In step S46, the vulnerability probabilities of each weak link in the optimal weak link set are calculated by the following formula: Among them, is the probability of the weak link of the node i ; is the true label of the node i ; = 1 indicates that the node i is a weak link, = 0 indicates that the node i is normal; is the probability that the node k given by the i th decision tree is a weak link; is the weight of the k th decision tree, K is the total number of decision trees.

10. The method for identifying weak links based on the power grid topology structure according to claim 2, wherein In step S5, a reward function is designed by comprehensively considering power supply reliability, power supply cost, and load uniformity, including: S51: The first optimization objective function is obtained with the goal of minimizing the sum of the power supply radii of each node in the power grid after adding sectional switches, where the positions of the added sectional switches are determined using the SA-PSO algorithm based on the identified weak links; the first optimization objective function is expressed as: Among them, is the power supply radius of each node in the power grid i ; x is the position where sectional switches are added; is the cost of adding sectional switches; is the weight for balancing the power supply radius and the switch cost; S52: According to the load distribution, the second optimization objective function is obtained with the goal of maximizing the balance of the loads of each transformer, expressed as: Among them, is the load of the transformer m ; is the total number of transformers; is the layout location of the transformer; S53: The third optimization objective function is obtained with the goal of the most balanced load distribution after the load transfer mechanism is activated; the load transfer mechanism is activated when the load on any line or node exceeds a preset threshold; the third optimization objective function is expressed as: Among them, represents the load of the line after load transfer n ; is the average load of all lines; is the selection variable of the load transfer path; S54: The first to third optimization objective functions are weighted and summed to obtain the reward function.

11. A weak link identification device based on the power grid topology structure adopts the weak link identification method based on the power grid topology structure as described in any one of claims 1-10, characterized in that, Including: A topology model establishment module for establishing a power grid topology model including nodes and edges, where the nodes include transformers, sectional switches, and load points, and the edges represent lines, and the power grid topology model contains the electrical parameters and geographical information of each node and edge; A weak link marking module for calculating the weak parameter corresponding to each load point, each line, and sectional switch during the historical period based on the power grid topology model, and marking the load points, lines, and sectional switches that do not meet the preset evaluation criteria as historical weak links according to the calculation results; A scoring model establishment module for establishing and training a weak link scoring model based on a graph neural network and deep learning based on the topology model and historical weak links, and extracting features from the real-time electrical parameters of the topology model to generate the real-time weak link scores of each node and edge; An identification module for dynamically identifying the weak links in the power grid based on the real-time weak link scores of each node and edge, using a genetic algorithm combined with local search and a random forest algorithm introducing a voting mechanism.

12. The weak link identification device based on the power grid topology structure according to claim 11, wherein, Also including: A training module for using the historical power grid state data and the identified weak links to design a reward function by comprehensively considering power supply reliability, power supply cost, and load uniformity, training a GAN network to generate various power grid optimization strategies, and training a DRL agent to perform priority ranking on the generated power grid optimization strategies and select the optimal combination of power grid optimization strategies; The various power grid optimization strategies include: adding sectional switches, optimizing the transformer layout, and adjusting the load transfer path; An optimization module for generating various possible real-time power grid optimization strategies using the trained GAN network according to the real-time power grid state, and performing priority ranking on the real-time power grid optimization strategies and selecting the optimal combination of real-time power grid optimization strategies using the trained DRL agent.

13. A terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1-10.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, the steps of the method according to any one of claims 1-10 are implemented.

Images