Intelligent power grid vulnerability analysis method and system

Abstract

The invention discloses an intelligent power grid vulnerability analysis method and system, and relates to the technical field of power grid analysis, and the method comprises the following steps: constructing an intelligent power grid graph structure based on a topological structure, business attributes and physical attributes of an intelligent power grid; performing graph representation learning to obtain a plurality of representation vectors of different dimensions; performing fusion, nonlinear transformation and mapping on the plurality of representation vectors to obtain a score of each node, performing descending sorting on different scores, and taking nodes corresponding to the first n scores as key nodes; different attack simulation evaluations are carried out on the key nodes, structural evaluation is carried out on the key nodes through vulnerability analysis indexes, and importance evaluation is carried out on the key nodes through service attributes; and weighting the first score, the second score and the third score to obtain a vulnerability score of the smart grid. According to the method, comprehensive analysis is carried out from multiple dimensions, and the comprehensiveness and accuracy of power grid vulnerability assessment are improved.

Description

Technical Field

[0001] This invention relates to the field of power grid analysis technology, and in particular to a method and system for analyzing the vulnerability of smart grids. Background Technology

[0002] Electricity is a fundamental infrastructure and vital guarantee for social progress and economic development. The safe and stable operation of power system networks is crucial for social stability and economic growth. Currently, power security has become a significant issue facing all countries worldwide and has been designated as a basic research area addressing major national strategic needs. With the large-scale interconnection of power grids and the increasing penetration of new energy sources such as wind and solar power, the complexity of power grid topology and dynamic performance is also increasing, making operation and control more difficult. Power systems are now evolving from isolated small grids to large-scale grids transmitting power over long distances, at ultra-high voltage, and even extra-high voltage. The network scale is becoming increasingly large and the structure increasingly complex. The complex network characteristics of large power grids are becoming more apparent, and the power system may face more threats, gradually increasing the possibility of large-scale power outages.

[0003] The complexity of power grids presents them with vulnerabilities and security challenges. When a critical node in the grid fails or is attacked, it can trigger a cascading failure, leading to widespread blackouts across the entire region. The nature of power outages is inextricably linked to the grid's network structure and operational characteristics; vulnerabilities in the grid structure are the underlying cause of major blackouts. Therefore, facing the dual pressures of new challenges and high requirements, it is urgent to focus on the security issues of complex power grids, identify critical nodes, assess and analyze grid vulnerabilities, and develop corresponding countermeasures.

[0004] In recent decades, scholars have conducted extensive research on the vulnerability and robustness of smart grids, proposing numerous vulnerability analysis methods. These mainly include: power grid vulnerability analysis methods based on mathematical statistics and probability theory, which primarily utilize statistics and probability theory to establish mathematical models for analysis; power grid vulnerability analysis methods based on complex network theory, which start from the structure of the power grid and reveal the complexity and vulnerability of the power grid system by analyzing the topology, the relationships between nodes, and the dynamic behavior of the network; power grid vulnerability analysis methods based on power system characteristics, which mainly analyze power grid vulnerability by judging the changes in node voltage amplitude and the degree of power flow exceeding the limits of transmission lines; and power grid vulnerability analysis methods based on intelligent optimization, which use improved ant colony optimization, simulated annealing, and particle swarm optimization, among other intelligent algorithms, for power grid vulnerability analysis.

[0005] While power grid vulnerability analysis methods based on mathematical statistics and probability theory have a strong theoretical foundation, they suffer from too many restrictive assumptions, high time complexity, slow operation, and low efficiency. Power grid vulnerability analysis methods based on complex network theory offer a more comprehensive and systematic framework, but these methods mostly assess vulnerability based on a single network topology, leading to low accuracy. Power grid vulnerability analysis methods based on power system characteristics focus solely on the analysis of these characteristics, resulting in low accuracy. The introduction of intelligent algorithms further improves the comprehensiveness of power grid vulnerability analysis, achieving better results, but the need for parameter setting and optimization, coupled with high time complexity, slow operation, and low efficiency, prevents them from meeting the needs of modern smart grids.

[0006] In summary, existing power grid vulnerability analysis methods suffer from low algorithm efficiency and low accuracy. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the present invention provides a method and system for vulnerability analysis of smart grids, which solves the existing problems.

[0008] The present invention adopts the following technical solution: In a first aspect, the present invention provides a method for vulnerability analysis of smart grids, comprising the following steps: Based on the topology, business attributes, and physical attributes of the smart grid, a smart grid graph structure is constructed, and the adjacency matrix and attribute matrix of each node in the smart grid graph structure are obtained. The topology, adjacency matrix, and attribute matrix of the smart grid are input into a graph representation model for graph representation learning, resulting in multiple representation vectors of different dimensions. These vectors are then fused, nonlinearly transformed, and mapped to obtain a score for each node. The scores are then sorted in descending order, and the top... n Each node corresponding to a score is designated as a key node; Different attack simulations are performed on key nodes to obtain the corresponding first score; vulnerability analysis indicators are established based on the smart grid topology, business attributes, node associations and dynamic behaviors, and the key nodes are structurally evaluated through the vulnerability analysis indicators to obtain the second score; the importance of key nodes is evaluated through business attributes to obtain the third score; the first, second and third scores are weighted to obtain the vulnerability score of the smart grid.

[0009] Preferably, the construction of the smart grid graph structure based on the topology, business attributes, and physical attributes of the smart grid specifically includes the following steps: The power grid topology is obtained by modeling the actual structure of the smart grid. The power transmission lines in the power grid topology are abstracted as edges of a complex network, and substations or power plants are abstracted as nodes of a complex network. The business attributes and physical attributes of the smart grid are integrated into the node attributes or edge attributes to construct a smart grid graph structure.

[0010] Preferably, the multiple representation vectors include node representation vectors, edge representation vectors, and subgraph representation vectors.

[0011] Preferably, multiple representation vectors are fused, nonlinearly transformed, and mapped based on a key node identification model, wherein the key node identification model includes a fusion layer, a graph neural network layer, and a fully connected layer.

[0012] Preferably, the training of the key node identification model specifically includes the following steps: The adjacency matrix of each node is input into the SIR model to obtain the corresponding importance score; A loss function is constructed by combining the score and importance score of each node, and the key node identification model is trained based on the loss function.

[0013] Preferably, the graph representation model adopts the GCN model.

[0014] Preferably, the vulnerability analysis metrics include node centrality, average path length, degree distribution, network efficiency, power flow betweenness, and electrical distance.

[0015] Preferably, the vulnerability score is as follows: ; in, Vulnerability score for each critical node, For second place, Third place score, For first place, , and These are the weighting coefficients.

[0016] Preferred options also include: Vulnerability warnings are issued based on vulnerability scores, and recovery strategies are provided from different dimensions.

[0017] Secondly, the present invention provides a smart grid vulnerability analysis system based on graph neural networks, comprising: The module is used to construct the smart grid graph structure based on the topology, business attributes and physical attributes of the smart grid, and to obtain the adjacency matrix and attribute matrix of each node in the smart grid graph structure. The identification module is used to input the topology, adjacency matrix, and attribute matrix of the smart grid into the graph representation model for graph representation learning, obtaining multiple representation vectors of different dimensions. These multiple representation vectors are then fused, nonlinearly transformed, and mapped to obtain a score for each node. The scores are then sorted in descending order, and the top... n Each node corresponding to a score is designated as a key node; The analysis module is used to perform different attack simulations and assessments on key nodes to obtain the corresponding first score; based on the smart grid topology, business attributes, node associations and dynamic behavior, vulnerability analysis indicators are established, and the key nodes are structurally assessed through the vulnerability analysis indicators to obtain the second score; the importance of key nodes is assessed through business attributes to obtain the third score; the first score, second score and third score are weighted to obtain the vulnerability score of the smart grid.

[0018] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects: This invention first constructs a smart grid graph structure based on the topology, business attributes, and physical attributes of the smart grid. The topology and grid system characteristics are integrated into the graph structure construction process. Based on this graph structure, key nodes in the smart grid are identified, improving the accuracy of key node identification. Finally, different attack simulations, structural assessments, and importance assessments are performed on the highly accurate key nodes. The corresponding scores are weighted and comprehensively analyzed from multiple dimensions to improve the comprehensiveness and accuracy of grid vulnerability assessment.

[0019] To address the efficiency issues arising from the expansion of power grid scale and structural complexity, this invention employs a graph representation learning method. This method associates and represents high-dimensional power grid structures and business-related data, and maps them to a low-dimensional vector space to achieve dimensionality reduction, thereby improving data processing efficiency. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 Changes in the network's vulnerability to attacks on critical edges and critical nodes; Figure 2 A multi-objective optimization architecture based on an improved genetic algorithm; Figure 3 This is a graph neural network structure based on the attention mechanism; Figure 4 This is a schematic diagram of a smart grid vulnerability analysis method according to the present invention; Figure 5 This is a schematic diagram of the power network construction according to the present invention; Figure 6 This is a flowchart of a traditional graph representation method. Figure 7 This is a flowchart of the graph depth characterization process of the present invention; Figure 8 This is a schematic diagram of the key node identification based on graph neural networks according to the present invention; Figure 9 This is a schematic diagram of the SIR propagation model of the present invention; Figure 10 This is a schematic diagram of the power network vulnerability analysis of the present invention; Figure 11 This is a schematic diagram of the worm-borne power grid attack of the present invention; Figure 12 This is a flowchart of a smart grid vulnerability analysis method according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Relevant background: Methods based on mathematical statistics and probability theory: These methods mainly rely on statistical analysis and probabilistic inference of power grid data to assess the vulnerability and potential risks of the power grid. (1) Statistical analysis of fault data. By collecting and analyzing power grid fault data, statistical indicators such as the frequency of occurrence, duration and impact range of different types of faults are obtained to identify fault modes that lead to power grid vulnerability. For example, statistical analysis can be performed on line faults, equipment faults, voltage imbalances, etc., to understand their occurrence patterns and possible consequences. (2) Probability inference of fault propagation. Based on fault data and power grid topology, probability inference methods are used to estimate the probability of fault propagation in the power grid. By analyzing factors such as the connectivity between power grid nodes, power transmission paths and load distribution of nodes, the probability of node faults on the fault propagation path is calculated, thereby identifying vulnerable links that may trigger cascading faults. Power grid vulnerability analysis methods based on mathematical statistics and probability theory have the advantages of solid theoretical foundation, data-driven and quantitative assessment, but these methods usually require a large number of assumptions, such as the independence, randomness or normal distribution of data, and cannot handle high-dimensional data, and have limitations in handling uncertainty.

[0024] Methods based on complex network theory: These methods primarily utilize the principles and methods of network science, treating the power grid as a complex network. By analyzing the network's topology, the relationships between nodes, and the network's dynamic behavior, they assess the power grid's vulnerability and potential risks. Figure 1 This demonstrates the structural changes of a network when it is attacked by critical edges and critical nodes. These methods, based on complex network theory, are significant for identifying vulnerable links, assessing risks, and developing vulnerability mitigation strategies. However, most of these methods judge the vulnerability of the power grid from a single network topology, neglecting the operational and physical attributes of the power grid; therefore, their analytical accuracy is not high.

[0025] Methods based on power system characteristics: These methods focus on the internal characteristics of the power grid, analyzing features such as power flow, short-circuit current, power stability, and power quality to identify vulnerable links and potential risks. However, these methods primarily analyze vulnerability from the perspectives of the power grid's physical and operational attributes, with insufficient consideration for network structure.

[0026] Intelligent optimization-based methods primarily utilize intelligent algorithms and optimization techniques to assess and analyze the vulnerability of power grids. These methods employ search and optimization algorithms to find the optimal performance and minimum vulnerability state of the power grid, providing corresponding vulnerability assessment results. For example... Figure 2 As shown, the optimal solution is searched by simulating gene crossover, mutation, and selection. Multi-objective population update strategies, such as multi-objective particle swarm optimization and multi-objective differential evolution, are employed to increase solution diversity and search the solution space. Most intelligent algorithms are based on traditional optimization methods, such as genetic algorithms, particle swarm optimization, and ant colony optimization. These methods require parameter setting and tuning, resulting in high time complexity, which can no longer meet the needs of increasingly complex large-scale interconnected power grids.

[0027] Graph Neural Network-Based Approaches: Graph Neural Networks (GNNs) are machine learning and deep learning methods used to process graph data. Compared to the limitations of traditional neural networks in processing structured data, GNNs can effectively handle graph data containing complex relationships and structures. Figure 3This paper demonstrates a graph neural network (GNN) structure based on an attention mechanism, mainly consisting of an input layer, hidden layers, and an output layer. The input layer includes the topology matrix and business attribute matrix of the power grid. The hidden layers typically contain multi-layer convolutional networks, specifically the attention module in this case. The output layer displays the features of the nodes. Graph neural network technology has been widely applied in various fields, such as infrastructure security fault analysis, social network analysis, product recommendation systems, and drug function prediction. It fully utilizes the structural information in graph data, enabling it to capture the relationships between nodes and perform efficient prediction and reasoning. Currently, research on graph neural networks is still in its early stages, with relatively few studies applying GNN technology to power grids. These studies mainly focus on operation and maintenance, load forecasting, and fault location. Due to the complexity of power grid structure and operations, there is currently little literature on using GNNs for power grid vulnerability analysis.

[0028] In summary, while power grid vulnerability analysis methods based on mathematical statistics and probability theory have a strong theoretical foundation, they suffer from too many assumptions and limitations, high time complexity, and inability to handle high-dimensional power grid data. Methods based on complex network theory provide a more comprehensive and systematic framework for power grid vulnerability analysis, but neglect the operational and physical attributes of the power grid. Methods based on power system characteristics focus only on the analysis of power grid system characteristics, with insufficient consideration for network structure. The introduction of intelligent algorithms further improves the comprehensiveness of power grid vulnerability analysis and can achieve good analytical results, but due to the need for parameter setting and optimization and high time complexity, it cannot meet the needs of modern smart grids. Methods based on graph neural networks consider both network structure and operational attributes, making them highly suitable for power grid vulnerability analysis, but they are still in their early stages, and no literature currently uses this technique to study power grid vulnerability.

[0029] (1) With the grid connection of new energy and microgrids, the power grid is becoming increasingly large and complex, which puts forward higher requirements for the running speed and accuracy of existing algorithms.

[0030] (2) Most existing methods only conduct single analysis of power grid structure or power grid business attributes, without integrating the combined impact of power grid structure and business attributes on power grid vulnerability.

[0031] (3) There is a lack of corresponding countermeasures for the vulnerability of the power grid and a lack of early warning for cascading failures caused by possible accidents. If a fault occurs at a vulnerable node in the power grid and is not controlled, the connection between the power grids will continuously amplify the impact of the fault, resulting in cascading failures and causing large-scale power outages in the regional power grid.

[0032] The above challenges are common problems faced in smart grid vulnerability analysis and are also difficulties in its application and implementation. To address these problems and challenges, this invention comprehensively analyzes relevant research results both domestically and internationally, and plans to use graph neural network methods to analyze smart grid vulnerabilities, while also researching corresponding countermeasures and providing early warnings. Details are as follows:

[0033] (1) To address the issues of algorithm efficiency and accuracy caused by the expansion of power grid scale and structural complexity, this invention adopts a graph representation learning method to associate and represent high-dimensional power grid structure and business data, and maps them to a low-dimensional vector space to achieve dimensionality reduction, thereby improving the efficiency of data processing and the accuracy of vulnerability analysis.

[0034] (2) In view of the problem of single analysis in traditional methods, this invention proposes a new smart grid vulnerability analysis method based on complex network and graph neural network technology. This method can integrate the structural characteristics of the power grid network and the characteristics of business attributes to identify key nodes in the smart grid and conduct comprehensive analysis from multiple dimensions.

[0035] (3) In response to the lack of countermeasures and early warnings for power grid vulnerabilities, this invention employs different attack strategies, such as critical node attacks, central node attacks, and large-scale node attacks, to analyze and study the robustness of the power network. Through these assessments and analyses, early warning information and corresponding countermeasures are provided for vulnerable links in the power grid.

[0036] This invention focuses on power grids, addressing the growing scale and complexity of power grids, which exacerbate their vulnerability. Based on complex network theory, it employs deep learning and graph neural network techniques to assess and analyze power grid vulnerability. First, the power grid is modeled, constructing a graph model based on its topology. Then, deep learning methods are used to represent the power grid. Next, a graph neural network-based method for identifying critical nodes in the power grid is proposed. Finally, a power grid vulnerability assessment model is constructed, employing various attack strategies to analyze power network vulnerabilities, revealing weaknesses, and providing early warnings and countermeasures. Figure 4 and Figure 12 As shown. Specifically, it includes the following steps:

[0037] S1: Graph Neural Network Modeling.

[0038] The power grid topology is modeled based on the actual structure of the smart grid to obtain the connection relationships between various components and lines. Based on the actual power grid topology and combined with complex network theory, transmission lines are abstracted as edges of a complex network, and substations or power plants are abstracted as nodes. The business and physical attributes of the smart grid are integrated into the node or edge attributes to construct a smart grid graph structure, obtaining the adjacency matrix and attribute matrix for each node. Based on the smart grid graph structure, the topological relationships and attribute characteristics between nodes are captured. Physical attributes describe the inherent, objectively existing engineering parameters and connection relationships of power grid components, such as electrical board parameters, equipment status, and topological roles. Electrical board parameters include voltage level, rated capacity, frequency, phase angle, short-circuit capacity, etc. Equipment status includes transformer load rate, circuit breaker opening / closing status, and fault probability, etc. Topological roles include node type and its hierarchy in the network.

[0039] Node types include generation, transmission, distribution, and load centers, and their network hierarchy includes backbone nodes and edge nodes. Business attributes describe the functional attributes derived from the operation, market, and management of the power grid, reflecting economic, control, and informational dimensions beyond power flow, such as load characteristics (load curves, power reliability requirements), control functions (whether it belongs to a dispatch center, automated substation, or distributed energy aggregation point), etc.

[0040] Based on the research data and foundation of previous power grid projects, and combined with complex network theory, a smart grid topology is established, and a real power network is constructed by combining the business attributes and physical attributes of the power network. Figure 5 This invention presents a schematic diagram of a smart grid model, showing a power grid network containing 39 nodes. It proposes to combine the LFR (Lancichinetti Fortunato Radicchi) model, the GN (Girvan Newman) model, and Python's networkx and igraph toolkits to design a network generation model. This model controls the network structure from multiple dimensions, including network size, degree distribution, average shortest path length, and clustering coefficient. It employs classic complex network models to construct different types of networks, such as random networks, scale-free networks, and small-world networks. Based on the business and physical attributes of real power networks, it establishes attributes and weights between nodes and edges within the network.

[0041] S2: Deep characterization of power networks.

[0042] To address the shortcomings of traditional methods in representing the structure and node attributes of power networks, this paper studies a representation method based on deep learning and graph neural networks. Utilizing graph representation learning, this method comprehensively considers the topology of the power grid, the operational attributes of grid nodes, and their physical attributes to perform a deep representation of the power grid's structure and attribute information. Nodes, lines, and attributes are transformed into low-dimensional vector representations for subsequent analysis and prediction.

[0043] The adjacency matrix is ​​extracted based on the power grid network structure. Combined with the power grid business attributes and physical attributes as input, the structural attributes and feature attributes of the power grid network are fused and represented by a graph deep learning model, and the output is a multi-dimensional representation at the node level, edge level and subgraph level.

[0044] The main goal of graph representation learning is to transform graph data into a low-dimensional, dense vectorized representation, while ensuring that the properties of the graph data also correspond in the vector space. Graph data representation can be at the node level or the entire graph level, and node representation learning has always been the primary focus of graph representation learning. Traditional graph representation learning is based on graph structure representation learning; the vector representations of nodes only come from the graph's topological structure and lack representations of the feature attributes of graph nodes or edges. Traditional graph representation learning methods, such as... Figure 6 As shown, deep representation learning based on graph features allows the vector representation of nodes to include not only the topological information of the graph but also other attribute feature vectors, such as the operational and physical attributes of the power grid. Figure 7 As shown, the adjacency matrix and attribute matrix of the power grid network are taken as input, and the graph depth representation model (GCN) is used to automatically aggregate the structural features and attribute features of the power grid. After nonlinear transformation of the model, low-dimensional vector representations of nodes, edges and subgraphs of different dimensions are output as needed. The adjacency matrix of the power grid network is represented by A, and the attribute feature matrix is ​​represented by H. When using it, the adjacency matrix needs to be transformed and then normalized, as shown in formula (1) and formula (2).

[0045] (1); in, The normalized adjacency matrix, for The identity matrix, It is usually set to 1.

[0046] (2); in, for The degree matrix, the main function of formula (2) is to normalize the data.

[0047] S3: Key Node Identification.

[0048] Research focuses on graph neural network models, such as Graph Convolutional Networks (GCN), GraphSAGE, and GAT. These models can utilize the connections between nodes during the learning process to transfer and aggregate node feature information, thereby obtaining a more comprehensive node representation.

[0049] We design a key power grid node identification model based on graph neural networks. This model aggregates neighbor information according to the network structure, automatically extracts the feature representation of nodes, and performs model training and optimization.

[0050] To achieve key node identification and evaluation, key nodes in the trained graph neural network model are identified and evaluated using evaluation metrics.

[0051] A key node identification model for power grids based on graph neural network technology is established, and related algorithms are designed. Using the representations of power grid nodes and subgraph representations as input, the key nodes in the power grid are accurately identified through model training and optimization. The key node identification model includes a fusion layer, a graph neural network layer, and a fully connected layer.

[0052] Graph Neural Networks (GNNs) are deep learning models used to process graph-structured data. They achieve learning and inference at the node and graph levels by extracting features from the local neighborhoods of nodes and edges. Compared to traditional key node identification methods, GNN-based key node identification methods have a more comprehensive ability to consider contextual information, handle complex graph structures, and possess advantages such as strong learning capabilities, robustness, and interpretability. This invention employs GNN technology to achieve key node identification in power grids. Figure 8 As shown, the node representation, edge representation, and subgraph representation obtained from the deep representation of the S2 power grid are used as inputs. After nonlinear transformation by graph neural networks (GCN, GAT, GraphSAGE), the features of the node itself and its surrounding neighbors are aggregated. The specific nonlinear transformation is shown in formula (3):

[0053] (3); in, For the first i Features of layer nodes It is a non-linear activation function, and the weights are shared among different nodes. This is the normalized adjacency matrix, which takes into account the importance of each node. For the first i The weights of a layered neural network, This is a bias term.

[0054] Next, the characteristics of each power grid node are output through the fully connected layer, and then... Logsoftmax Calculate the score for each node: (4); in, Represents nodes in the network i , This involves summing the exponent values ​​of all nodes in the network for the purpose of normalization. log It is the natural logarithm.

[0055] Finally, the ranking results are compared with those obtained by the SIR (Susceptible-Infected-Recovered) model. The loss is calculated, and the graph neural network is iteratively optimized based on the loss. The input of the SIR model is the adjacency matrix of the network, and the output is the number of infected nodes for each node. This number is normalized and used as the importance score for each node. The nodes are then sorted in descending order of their scores to obtain the ranking results. Its propagation model is shown in the following equation:

[0056] (5); in, , and They represent The number of susceptible nodes, infected nodes, and recovered nodes at any given time. Parameters Indicates the probability of infection. This represents the recovery probability. Assume the total number of nodes in the network is... The total number of nodes in the network is as shown in formula (6).

[0057] (6); The SIR model, as a general metric, is used to evaluate the importance of nodes in a network. In this model, the letters S, I, and R represent Susceptible, Infected, and Recovered, respectively, and their dynamic effects are as follows: Figure 9 As shown.

[0058] The loss is calculated as follows: (7); in, Indicates the number of nodes. It is a node The true value is obtained from the SIR model. It is a node The predicted value is obtained from the key node identification model.

[0059] Through continuous iterative optimization of the model, the calculated loss decreases. When the loss reaches a set threshold, the importance of each node in the network can be determined. The top n nodes in the network are then identified as critical nodes, with the value of n set according to the specific application scenario. Finally, the learned model is applied to a real power grid to identify critical nodes.

[0060] S4: Vulnerability Analysis, Early Warning, and Countermeasures.

[0061] Vulnerability analysis indicators are established based on the power grid topology, business attributes, node relationships, and dynamic behavior. Vulnerability analysis methods are designed by combining complex network theory and power grid characteristics, and an evaluation model is built to conduct vulnerability analysis and evaluate the power grid. Then, the robustness of the power network is measured using different attack methods, such as random attacks, large-degree node attacks, intermediate node attacks, and critical node attacks, and early warning information is provided. Countermeasures are offered from different perspectives, including critical node recovery, target recovery, important link recovery, and dependency recovery.

[0062] Traditional methods for studying power grid node vulnerability indices rely on single indicators of electrical characteristics or network structure, such as electrical distance, node degree, or betweenness coefficient. However, as power grids grow larger and more complex, these single-indicator analyses are no longer sufficient to reflect the electrical characteristics of large power grids. Therefore, it is necessary to develop vulnerability indices that better reflect real-world power grid models. Power network vulnerability is related not only to the network's topology and physical attributes but also to the precision of attack methods. Therefore, this invention establishes analytical indices from different perspectives, including topology, service attributes, inter-node correlations, and related dynamic behaviors of nodes. These indices include node centrality, average path length, degree distribution, network efficiency, power flow betweenness coefficient, and electrical distance.

[0063] The power grid vulnerability assessment model studies the use of vulnerability indicators to analyze and evaluate the identified key nodes in the power grid. By analyzing multiple indicators such as network topology, service attributes, node correlation and dynamic behavior, the power grid vulnerability assessment model is constructed to assess the vulnerability of the power grid.

[0064] This research focuses on power grid attack simulation, targeting critical nodes, intermediate nodes, and large nodes within the power grid. By simulating various attack or interference scenarios, the study analyzes the power grid's response and recovery capabilities, assesses its vulnerability, and evaluates its ability to maintain stable operation.

[0065] The study proposes corresponding countermeasures and early warnings, provides power grid enhancement strategies based on the power grid vulnerability analysis results, and provides early warning information based on the attack simulation results.

[0066] Based on complex network theory, statistical learning, and machine learning, this study employs multiple methods, including structural analysis, operational analysis, attack analysis, and comprehensive analysis, to analyze the vulnerability of the power grid. An evaluation model is then constructed to assess the vulnerability of each key node, yielding vulnerability scores. (8); in, The structural analysis score of key nodes in the power grid is obtained through degree, betweenness, and eigenvector centrality indices. It represents business attributes (such as load and traffic), assesses the functional importance of key nodes, and is obtained through power grid data collection; This represents the score obtained by attacking critical nodes in the power grid. Specifically, the score is calculated by removing the critical nodes and counting the number of subgraphs in the network. , , The weighting coefficients are determined through machine learning, or they can be set based on expert experience.

[0067] Then, vulnerability warnings are issued based on the obtained assessment results; finally, recovery strategies are proposed from different dimensions. Power grid vulnerability analysis research, such as... Figure 10 As shown, Figure 11 This demonstrates a worm-based attack on a power grid network, where S represents a susceptible node, I represents an infected node, and R represents an isolated node.

[0068] Based on the same concept, the present invention also provides a smart grid vulnerability analysis system based on graph neural networks, including a construction module, an identification module and an analysis module.

[0069] The construction module is used to build a smart grid graph structure based on the topology, business attributes and physical attributes of the smart grid, and to obtain the adjacency matrix and attribute matrix of each node in the smart grid graph structure.

[0070] The identification module is used to input the topology, adjacency matrix, and attribute matrix of the smart grid into the graph representation model for graph representation learning, obtaining multiple representation vectors of different dimensions. These multiple representation vectors are then fused, nonlinearly transformed, and mapped to obtain a score for each node. The scores are then sorted in descending order, and the top... n Each node corresponding to a score is designated as a key node.

[0071] The analysis module is used to perform different attack simulations and assessments on key nodes to obtain the corresponding first score; based on the smart grid topology, business attributes, node associations and dynamic behavior, vulnerability analysis indicators are established, and the key nodes are structurally assessed through the vulnerability analysis indicators to obtain the second score; the importance of key nodes is assessed through business attributes to obtain the third score; the first score, second score and third score are weighted to obtain the vulnerability score of the smart grid.

[0072] This project focuses on smart grids, addressing their vulnerabilities. It utilizes graph representation methods and complex network theory to extract and characterize the intrinsic connections and node features between grid nodes; establishes a key node identification model based on graph neural networks; integrates evaluation indicators to construct an assessment model that reveals vulnerable nodes in the grid; and provides early warnings and robustness enhancement strategies through simulated attacks.

[0073] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0074] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A vulnerability analysis method for smart grids, characterized in that, Includes the following steps: Based on the topology, business attributes, and physical attributes of the smart grid, a smart grid graph structure is constructed, and the adjacency matrix and attribute matrix of each node in the smart grid graph structure are obtained. The topology, adjacency matrix, and attribute matrix of the smart grid are input into the graph representation model for graph representation learning, resulting in representation vectors with multiple dimensions. Multiple representation vectors are fused, nonlinearly transformed, and mapped to obtain the score for each node. The different scores are then sorted in descending order, and the top... n Each node corresponding to a score is designated as a key node; Different attack simulations were performed on key nodes to evaluate their performance and obtain the corresponding first score. Vulnerability analysis indicators are established based on the smart grid topology, business attributes, node associations, and dynamic behavior. The structural assessment of key nodes is conducted using these indicators to obtain a second score. The importance assessment of key nodes is conducted using business attributes to obtain a third score. The first, second, and third scores are weighted to obtain the vulnerability score of the smart grid.

2. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The construction of the smart grid graph structure based on the topology, business attributes, and physical attributes of the smart grid specifically includes the following steps: The power grid topology is obtained by modeling the actual structure of the smart grid. The power transmission lines in the power grid topology are abstracted as edges of a complex network, and substations or power plants are abstracted as nodes of a complex network. The business attributes and physical attributes of the smart grid are integrated into the node attributes or edge attributes to construct a smart grid graph structure.

3. The smart grid vulnerability analysis method as described in claim 1, characterized in that, Multiple representation vectors include node representation vectors, edge representation vectors, and subgraph representation vectors.

4. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The key node identification model is used to fuse, nonlinearly transform, and map multiple representation vectors. The key node identification model includes a fusion layer, a graph neural network layer, and a fully connected layer.

5. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The training of the key node identification model specifically includes the following steps: The adjacency matrix of each node is input into the SIR model to obtain the corresponding importance score; A loss function is constructed by combining the score and importance score of each node, and the key node identification model is trained based on the loss function.

6. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The graphical representation model adopted is the GCN model.

7. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The vulnerability analysis metrics include node centrality, average path length, degree distribution, network efficiency, power flow betweenness, and electrical distance.

8. The smart grid vulnerability analysis method as described in claim 1, characterized in that, The specific vulnerability scores are as follows: ； in, Vulnerability score for each critical node, For second place, Third place score, For first place, , and These are the weighting coefficients.

9. The smart grid vulnerability analysis method as described in claim 1, characterized in that, Also includes: Vulnerability warnings are issued based on vulnerability scores, and recovery strategies are provided from different dimensions.

10. A smart grid vulnerability analysis system based on graph neural networks, characterized in that, include: The module is used to construct the smart grid graph structure based on the topology, business attributes and physical attributes of the smart grid, and to obtain the adjacency matrix and attribute matrix of each node in the smart grid graph structure. The identification module is used to input the topology, adjacency matrix, and attribute matrix of the smart grid into the graph representation model for graph representation learning, obtaining multiple representation vectors of different dimensions. These representation vectors are then fused, nonlinearly transformed, and mapped to obtain a score for each node. The scores are then sorted in descending order, and the top... n Each node corresponding to a score is designated as a key node; The analysis module is used to perform different attack simulations and evaluations on key nodes to obtain the corresponding first score; Vulnerability analysis indicators are established based on the smart grid topology, business attributes, node associations, and dynamic behavior. The structural assessment of key nodes is conducted using these indicators to obtain a second score. The importance assessment of key nodes is conducted using business attributes to obtain a third score. The first, second, and third scores are weighted to obtain the vulnerability score of the smart grid.

Images