Power distribution network communication network vulnerability assessment method based on graph neural network

Abstract

The invention relates to the technical field of power distribution network operation and maintenance, in particular to a power distribution network communication network vulnerability assessment method based on a graph neural network, comprising the following steps: acquiring power distribution network communication network operation data including node equipment attribute data and link attribute data; based on the operation data of the communication network of the power distribution network, grading and marking the vulnerability of each node to form a training data set; establishing a power distribution network communication network vulnerability evaluation model; and training a power distribution network communication network vulnerability evaluation model by taking the operation data in the training data set as input and the vulnerability level label as supervision. According to the embodiment of the invention, the operation data of the power distribution network communication network part or the historical nodes are labeled in advance, and then the trained model is used to comprehensively and accurately evaluate the vulnerability level of each node in the power distribution network communication network.

Description

Technical Field

[0001] The present application relates to the technical field of distribution network operation and maintenance, and in particular to a distribution network communication network vulnerability assessment method based on graph neural network. Background Art

[0002] In the existing technology, the distribution network is relatively large in scale, relatively complex in structure, and has high internal data flow transmission requirements, which makes the distribution network communication network highly vulnerable when facing special circumstances. Once the communication network fails, it may lead to uneven power distribution, power outages, and even more serious economic losses and social impacts.

[0003] Therefore, a method is needed to identify the vulnerability of the distribution network and to achieve a rapid, accurate and comprehensive assessment of each node under the distribution network communication network, so that operators can handle the vulnerable nodes of the distribution network communication network. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present application provides a distribution network communication network vulnerability assessment method based on graph neural network. The embodiment of the present application is to use the operating data of part of the distribution network communication network or historical nodes to pre-label, and then use the trained model to comprehensively and accurately assess the vulnerability level of each node in the distribution network communication network.

[0005] The above application objectives of this application are achieved through the following technical solutions:

[0006] A distribution network communication network vulnerability assessment method based on graph neural network includes the following steps:

[0007] Acquire the operation data of the distribution network communication network, the operation data including node device attribute data and link attribute data;

[0008] Based on the operation data of the distribution network communication network, the vulnerability of each node is graded and labeled to form a training data set;

[0009] Establish a distribution network communication network vulnerability assessment model;

[0010] The distribution network communication network vulnerability assessment model is trained using the operating data in the training dataset as input and the vulnerability level labels as supervision;

[0011] Obtain the current distribution network communication network operation data;

[0012] The distribution network communication network vulnerability assessment model is used to grade the vulnerability of the current distribution network communication network nodes.

[0013] Optionally, the vulnerability of each node is graded and marked, including:

[0014] Dividing a number of nodes under the distribution network into a first set and a second set;

[0015] Obtain the vulnerability level label of each node in the first set evaluated by experts;

[0016] A vulnerability level label is assigned to each node in the second set according to the centrality index of each node in the second set.

[0017] Optionally, a vulnerability level label is assigned to each node in the second set according to the centrality index of each node in the second set:

[0018] According to the weights of the centrality indicators of the nodes in the second set, a Shannon entropy weighted combination centrality indicator is established, wherein the centrality indicators of the nodes include degree centrality, betweenness centrality, node centrality and node reliability;

[0019] Calculate the centrality index of the nodes in the second set based on the Shannon entropy weighted combination centrality index;

[0020] The vulnerability levels of the nodes in the second set are marked according to the centrality indexes of the nodes in the second set.

[0021] Optionally, the establishing of the Shannon entropy weighted combination centrality index includes:

[0022] Get the centrality index parameters of the node, the centrality index parameters of the node include degree centrality C D , betweenness centrality C B , node centrality C P and node reliability C R ;

[0023] Determine degree centrality, betweenness centrality, node centrality and entropy measure E of node reliability D 、E P 、E B 、E R ;

[0024] Determine the weight w of each centrality index of the node i ;

[0025] Obtain the Shannon entropy weighted combination centrality index:

[0026] C=w 1 C D +w 2 C P +w 3 C B +w 4 C R .

[0027] Optionally, the distribution network communication network vulnerability assessment model has a graph attention neural network extraction and decision tree ensemble learning mechanism.

[0028] Optionally, the establishing of a distribution network communication network vulnerability assessment model includes:

[0029] Establish a graph neural network model for extracting distribution network communication network features;

[0030] Determine the objective function;

[0031] Based on the objective function, a decision tree is constructed layer by layer.

[0032] Optionally, the step of grading the vulnerability of the current distribution network communication network nodes using the distribution network communication network vulnerability assessment model includes:

[0033] Extracting the communication network feature vector of the distribution network;

[0034] Based on the feature vector, the vulnerability level of the node is calculated using a decision tree.

[0035] Optionally, extracting the distribution network communication network feature vector includes:

[0036] Extracting node feature vectors of each node;

[0037] Extracting link feature vectors of each link;

[0038] Establish node weight matrix W v and the edge weight matrix W e ;

[0039] According to the node weight matrix W v and the edge weight matrix W e Perform linear transformation on node feature vectors and link feature vectors respectively;

[0040] Calculate the attention coefficient of each node;

[0041] According to the attention coefficient of each node, the feature information of the node neighbors is aggregated.

[0042] In summary, this application has the following beneficial technical effects:

[0043] The embodiment of the present application uses the operation data of the distribution network communication network part or the historical nodes to be marked in advance, and then uses the trained model to comprehensively and accurately evaluate the vulnerability level of each node in the distribution network communication network. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of an implementation method of the present application. DETAILED DESCRIPTION

[0045] The following is combined with Figure 1 This application is described in further detail.

[0046] In order to more clearly understand the technical solutions presented in the embodiments of the present application, a brief introduction is first given to the operation and evaluation of the distribution network communication network in the prior art.

[0047] In the new power system, the stability and security of the distribution network play an increasingly prominent role in supporting social and economic activities. With the widespread access of new power facilities such as distributed power sources, microgrids and virtual power plants, the scale of the distribution network has become larger, the structure has become more complex, and the transmission demand for data flow and control instructions has increased dramatically. This complexity not only increases the vulnerability of the distribution network in the face of natural disasters or man-made attacks, but also makes the stability of its communication network the key to ensuring the reliable operation of the power grid. The vulnerability of the distribution network communication network directly affects the stability and security of power supply, which is particularly important for critical infrastructure such as hospitals, schools, and commercial centers. Once the communication network fails, it may lead to uneven power distribution, power outages, and even more serious economic losses and social impacts. Due to the large-scale network, the complexity of the network structure, the diversity of equipment, and the openness of the network environment, it is particularly important to be able to efficiently handle large-scale networks and conduct a comprehensive and accurate network vulnerability assessment.

[0048] At present, there are mainly three methods for network vulnerability assessment:

[0049] Vulnerability assessment method based on static topology: The vulnerability of the network is assessed by analyzing the structural characteristics of the network. This method focuses on the configuration of nodes and connections in the network, and uses graph theory indicators such as node degree, clustering coefficient, betweenness, etc. to assess the vulnerability of important nodes or connections in the network. By identifying these key nodes or paths, it is possible to predict the possible performance of the network when facing attacks or failures. However, this method focuses on the static structural characteristics of the network and may ignore the properties of network devices, network traffic, dynamic changes, and the impact of actual operating conditions. In addition, this method assumes that all connections in the network have the same importance and fails to take into account the actual functions and load differences of each node and connection in the network. Therefore, it may not accurately reflect the vulnerability of the network in actual operation.

[0050] Vulnerability assessment method based on cascading failures: This method assesses the chain reactions that may occur in the network after an initial failure. This method simulates the failure of one or more network elements (such as substations in the power grid or routers in the communication network) and observes how these failures propagate through the network to cause a wider impact. This method focuses on the resilience of the network to initial failures, which can reveal the potential weaknesses of the network and the overall stability of the system. It is suitable for real-time monitoring and early warning of complex networks. This method is highly sensitive to the selection of the initial fault point when simulating the fault propagation process, and its assessment results may vary greatly depending on the selection of different initial fault points. In addition, the computational complexity of this method increases significantly with the increase in the size of the network, which may make it difficult to implement in large-scale networks. The accuracy of the cascade model also relies on assumptions and simplifications of network behavior, which may not fully conform to reality.

[0051] Vulnerability assessment method based on attack simulation: evaluates the vulnerability of the network by simulating external attacks or internal failures. This method usually uses software tools to generate simulated attack scenarios, such as DDoS attacks, malware propagation, etc., and then observes the response and recovery capabilities of the network. Attack simulation can help network administrators understand how the network performs when facing real-world threats, thereby optimizing defense strategies and emergency response plans. This method provides a means of actual combat exercises to help improve network security. Relying on preset attack scenarios, these scenarios may not cover all potential threat types, especially emerging or unknown attack methods. In addition, this method may require significant resource investment, including advanced security expertise and complex software tools. The results of attack simulation may also be too dependent on the settings and assumptions of the simulation environment, so there may be deviations in predicting actual attacks.

[0052] In general, in the current new power system distribution network, the static topology-based method cannot adapt to diverse and complex equipment, ignores its actual functions and load differences, and affects the accuracy of the evaluation; the cascading failure-based method cannot cope with large-scale networks, affecting its evaluation efficiency; the attack simulation-based method relies on preset scenarios and cannot discover and evaluate potential and unknown threats and vulnerabilities in the face of an open network environment.

[0053] In order to solve the above technical problems, the embodiment of the present application provides a distribution network communication network vulnerability assessment method based on graph neural network, comprising the following steps:

[0054] S101, obtaining distribution network communication network operation data, the operation data including node device attribute data and link attribute data;

[0055] S102, according to the distribution network communication network operation data, the vulnerability of each node is graded and labeled, and the distribution network communication network operation data is in a corresponding relationship with the vulnerability level, thereby forming a training data set;

[0056] S103, establishing a distribution network communication network vulnerability assessment model;

[0057] S104, using the operating data in the training data set as input and the vulnerability level label as supervision to train a distribution network communication network vulnerability assessment model;

[0058] S105, obtaining current distribution network communication network operation data, including node device attribute data and link attribute data;

[0059] S106. Use the trained distribution network communication network vulnerability assessment model to classify the vulnerability of the current distribution network communication network nodes, and output the current distribution network communication network node vulnerability level.

[0060] The following is a further introduction based on specific usage scenarios.

[0061] First, step S101 is executed. In simple terms, the distribution network communication network is composed of device nodes and communication links between device nodes. In the network topology structure, the aforementioned links can also be represented as edges. The distribution network communication network can be represented by a graph G = (V, E), where the network topology defined by the node V and the edge E provides insights into the dynamic operation of the power grid. The nodes in the data set represent the communication components on devices such as substations, power plants, and smart meters, and the edges represent the operation links and data paths that interconnect these nodes. First, the operation data, node device attribute data, and network structure attribute data under the network are collected;

[0062] Among them, the attributes of the node include: device asset type, device confidentiality level, device storage space, device delay tolerance, device data flow intensity, device data flow clustering coefficient, etc. The network structure information includes the network node set V and link attribute data. The link attribute data includes the network edge set E, edge type, edge operation frequency, edge length, edge congestion degree, edge data flow bandwidth, etc.

[0063] Then, step S102 is performed. The nodes of the communication network may include device nodes that have had problems in the past, key transmission nodes, or new, unverified nodes. Exemplarily, the vulnerability level Y marked in the embodiment of the present application may include "low", "medium", "high", and "severe". According to the distribution network communication network operation data, the vulnerability of each node is graded and marked. The distribution network communication network operation data corresponds to the vulnerability level, so as to form a training data set. The aforementioned distribution network communication network operation data may be historical data.

[0064] In step S103, a distribution network communication network vulnerability assessment model for the distribution network communication network is established, the assessment model takes the operation data of the distribution network communication network as input, and the output result is the vulnerability level of each node under the distribution network;

[0065] Then, step S104 is executed to train a distribution network communication network vulnerability assessment model using the operating data in the training data set as input and the vulnerability level label as supervision;

[0066] Then, step S105 is executed to obtain the current distribution network communication network operation data, including node device attribute data and link attribute data;

[0067] S106. Use the trained distribution network communication network vulnerability assessment model to classify the vulnerability of the current distribution network communication network nodes, and output the current distribution network communication network node vulnerability level.

[0068] In general, the embodiments of the present application use the operating data of the distribution network communication network part or historical nodes to be annotated in advance, and then use the trained model to comprehensively and accurately evaluate the vulnerability level of each node in the distribution network communication network.

[0069] As a feasible specific implementation method of the embodiment of the present application, the hierarchical labeling of the vulnerability of each node includes the following steps:

[0070] Step S201: dividing a number of nodes in the distribution network into a first set and a second set. In some possible implementations of the embodiment of the present application, some key nodes in the distribution network may be divided into the first set, and ordinary nodes may be divided into the second set;

[0071] Step S202: First, a portion of key nodes, i.e., several nodes in the first set, may be selected for manual labeling by domain experts. Of course, historical operation data of some nodes and expert evaluation levels under the operation data may also be obtained, thereby obtaining vulnerability level labels of some nodes evaluated by experts.

[0072] Step S203: assign vulnerability level labels to the nodes in the distribution network other than the nodes evaluated by experts according to the centrality indexes of the nodes in the second set.

[0073] Through the method of expert mixed annotation, both the quality and efficiency of annotation are improved.

[0074] The steps of assigning vulnerability level labels to the nodes in the second set according to the centrality indexes of the nodes in the second set include:

[0075] Step S2031: Obtain the centrality index parameters of the node, the centrality index parameters of the node include degree centrality C D , betweenness centrality C B , node centrality C P and node reliability C R , the aforementioned nodes are nodes in the second set;

[0076] Degree centrality C D It represents the number of adjacent nodes connected to the node, reflecting the direct influence of the node, where deg(v) is the degree of node v and |N| is the total number of nodes, which is calculated as follows:

[0077]

[0078] Node centrality C P (PageRank Centrality) evaluates the importance of a node based on the quality and quantity of its links, where B v is the set of nodes linked to v, L(u) is the number of links from node u, and d is the damping factor, which is calculated as follows:

[0079]

[0080] Betweenness centrality (C B ) represents the frequency of a node acting as a bridge between adjacent nodes in the network, reflecting the key role of nodes in information dissemination, where σ st is the total number of shortest paths from node s to node t, σ st (v) is the total number of paths that pass through v, which is calculated as follows:

[0081]

[0082] Node reliability (C R ) The importance of a node is measured by the change in the number of spanning trees after deleting the node. The greater the change, the more important the node is. τ(Gv) refers to the number of spanning trees in the network after deleting node v; τ(G) refers to the number of spanning trees when the network is complete, which is calculated as follows:

[0083]

[0084] Step S2032: The above four indicators are difficult to fully reflect all aspects of the importance of the node alone. To overcome this limitation, the embodiment of the present application adopts the Shannon entropy weighted combination centrality method to combine multiple indicators into one score to more comprehensively evaluate the vulnerability and importance of the nodes in the network. The weight of each indicator is determined by using Shannon entropy, aiming to quantify the inherent uncertainty and information diversity of the network nodes.

[0085] The Shannon entropy formula is used to quantify the diversity and uncertainty of each indicator, thereby assigning it a corresponding weight. The Shannon entropy of a discrete random variable X is composed of the possible values ​​of the indicator {x 1 , x 2 , ..., x n} and the probability density function p(x) that defines the proportion of nodes in the network with this specific indicator value x:

[0086]

[0087] The weight w of each indicator is calculated by the Shannon entropy of each indicator i , i is its index, where the entropy measure of degree centrality, betweenness centrality, PageRank, and node reliability is E D 、E P 、E B 、E R , thus, the entropy measure E of degree centrality, betweenness centrality, node centrality and node reliability is determined D 、E P 、E B 、E R ;

[0088] Step S2033: Determine the weight w of each centrality index under the node i , weight w i The calculation of is as follows:

[0089]

[0090] Step S2034: According to the weights of each centrality index of the node, a Shannon entropy weighted combination centrality index is established:

[0091] C=w 1 C D +w 2 C P +w 3 C B +w 4 C R ;

[0092] Step S2035: Calculate the centrality index of the node based on the Shannon entropy weighted combination centrality index. For example, the Shannon entropy weighted combination centrality index of each node is calculated by the formula in step S3024, and then the vulnerability level is divided into severe, high, medium, and low. The top 5% is marked as "severe", 5% to 20% is marked as "high", 20% to 75% is marked as "medium", and 75% to 100% is marked as "low".

[0093] The above hybrid labeling method can combine the in-depth understanding of human experts on network vulnerable nodes and the centrality indicator measurement of various dimensions to achieve efficient and accurate assessment of node vulnerability in smart grids. By using centrality indicators to automatically generate labels to expand the samples manually labeled by experts, it can improve labeling efficiency while maintaining high labeling quality.

[0094] It can be found that the expert-Shannon entropy hybrid labeling method of network node vulnerability combines the professional experience of human experts with the automatic labeling pseudo-labels of the Shannon entropy indicator, efficiently expanding the labeled data set of communication node vulnerability in the distribution network. Compared with the traditional manual labeling method, it can more efficiently process the data of large-scale communication system networks, has stronger adaptability and scalability, and helps deep learning model evaluation to improve accuracy and generalization.

[0095] As a feasible specific implementation method of the embodiment of the present application, the distribution network communication network vulnerability assessment model GraphBoost has a graph attention neural network extraction and decision tree ensemble learning mechanism. Specifically, the distribution network communication network vulnerability assessment model GraphBoost in the embodiment of the present application uses a graph attention neural network to extract the vulnerability characteristics in the distribution network communication network, and then performs vulnerability assessment on the nodes in the network by decision tree ensemble learning based on extreme gradient boosting.

[0096] The establishment of the distribution network communication network vulnerability assessment model GraphBoost in the embodiment of the present application may include the following steps:

[0097] Step S301: Establish a graph neural network model for extracting the characteristics of the distribution network communication network. The graph attention neural network is used to process the graph structure data in the distribution network communication network. It introduces the attention mechanism to learn the dependency relationship between device nodes, so as to effectively represent and process the node characteristics and edge characteristics in the graph structure data. The algorithm flow mainly includes initialization, feature transformation, calculation of attention coefficient, neighbor information aggregation and multi-head attention mechanism;

[0098] Step S302: Determine the objective function. In order to improve the accuracy and efficiency of node vulnerability assessment and facilitate expansion to new environments for assessment, a decision tree ensemble learning method based on extreme gradient boosting (XGBoost) is adopted. This ensemble learning method uses the feature vector obtained by GAT as input, effectively processes multi-dimensional feature input, and models complex patterns in the data to obtain the node vulnerability assessment result y∈{"low", "medium", "high", "severe"}. The objective function consists of a loss function and a regular term. The objective function is as follows:

[0099]

[0100] Among them, y i Node v i The actual vulnerability level label, is the vulnerability level label predicted by the model, and f k is the mth decision tree, L() is the loss function, Ω() is the regularization term, which usually includes the number of leaf nodes in the tree and the L2 norm of the leaf node weights. In order to optimize the objective function, the first-order and second-order Taylor expansions are used to approximate the loss function:

[0101]

[0102] Among them, g i and g′ i They are respectively the loss function for the predicted value This allows the algorithm to effectively use the gradient (g i ) and the Hessian matrix (g′ i ) to guide the growth of decision trees;

[0103] Step S303: Based on the objective function, a decision tree is constructed layer by layer from top to bottom using a greedy algorithm. The best segmentation method for each node is determined by maximizing the gain brought by the segmentation:

[0104]

[0105] Among them, I, I L and I R Represent the sample sets of the parent node and the left and right child nodes after the split of the decision tree respectively; λ and γ are regularization parameters, corresponding to the L2 regularization of the leaf node weight and the complexity penalty added to each leaf, respectively. In the integration process of multiple trees, the feature vector is used to evaluate the node vulnerability when multiple decisions are made, and a learning rate η is used to control the contribution of each tree to the final prediction to enhance the robustness of the model:

[0106]

[0107] Finally, the prediction results of all trees are added together to form the final node vulnerability assessment result y∈{“low”, “medium”, “high”, “severe”}. In this process, the method considers the contribution of each feature of the node feature vector to the final vulnerability classification and dynamically adjusts the learning focus.

[0108] The graph decision tree model (GraphBoost) proposed in the embodiment of the present application aims to deeply analyze the graph structure of the distribution network communication network and traditional network performance indicators, reliability indicators, etc. through the graph attention network (GAT), learn and extract the deep-level relationships and features of the nodes, and then use the XGBoost ensemble learning model to further adapt to the dynamic changes of the feature distribution to improve the accuracy and generalization of network node evaluation. Compared with traditional single models or direct statistical methods, the graph decision tree model can more comprehensively utilize the intrinsic connections of graph data and its interaction effects with other data to provide more accurate prediction results. In addition, this method combines the ability of GAT in capturing complex network dynamics and the efficiency of XGBoost in processing large-scale data sets, so that in a dynamically changing distribution network environment, it can evaluate and respond to the vulnerability of the network in real time and accurately, thereby optimizing network operations and enhancing the stability and security of the system.

[0109] Simply put, the graph attention-based integrated learning vulnerability assessment model GraphBoost proposed in the embodiment of the present application combines heterogeneous data of device node attributes and communication link edge attributes to achieve a more accurate and comprehensive vulnerability assessment. Compared with traditional single models or direct statistical methods, the embodiment of the present application can more comprehensively utilize the inherent connections of graph data and its interaction effects with other data, provide more accurate assessment results, accurately assess and respond to network vulnerabilities, and improve the stability and security of the system network.

[0110] After the distribution network communication network vulnerability assessment model GraphBoost is established and trained, the model can be used to assess the vulnerability of each node in the distribution network. The specific steps are as follows:

[0111] Step S601: Extract the characteristic vector of the distribution network communication network, including extracting the node characteristic vector of each node and the link characteristic vector of each link. The distribution network communication network can be represented by a graph G = (V, E), where V is a set of device nodes, E is a set of edges of operation links and data paths, and each node v has an initial characteristic vector x v , including device asset type, device confidentiality level, device storage space, device delay tolerance, device data flow intensity, device data flow clustering coefficient, etc. Each edge e ij =(v i ,v j ) has an initial eigenvector It includes edge type, edge operation frequency, edge length, edge congestion level, edge data flow bandwidth, etc.

[0112] Step S602: Establish a learnable node weight matrix W v and the edge weight matrix W e ;

[0113] Step S603: According to the node weight matrix W v and the edge weight matrix W e Perform linear transformation on node feature vectors and link feature vectors respectively:

[0114]

[0115] Among them, h v is the transformed node feature vector, is the transformed edge eigenvector;

[0116] Step S604: Calculate the attention coefficient of each node, and dynamically calculate the importance of each edge through the attention mechanism. i and node v j The edge between them has an attention coefficient α ij The calculation is as follows:

[0117]

[0118] Among them, a is a learnable attention vector, [·∥·∥·] represents the concatenation of vectors, LeakyReLU is a nonlinear activation function, and the softmax function is used for all connections to node v j The attention coefficient of the edge is normalized:

[0119]

[0120] in, Represents node v i The set of neighbor nodes;

[0121] Step S605: According to the attention coefficient of each node, aggregate the feature information of the node neighbors:

[0122]

[0123] where σ is the activation function, such as ReLU. is node v i Updated feature vector of ;

[0124] Step S606: In order to enhance the capability and stability of the model, GAT usually adopts a multi-head attention mechanism, in which K independent attention mechanisms run in parallel, and the feature vector output by each head is averaged:

[0125]

[0126] in, is the feature vector calculated by the kth attention mechanism head. Finally, is the extracted node v i The node vulnerability feature vector is obtained and the node vulnerability level is calculated using the decision tree.

[0127] According to the topological structure of the distribution network communication network, the model uses the attribute data of nodes and edges to learn the comprehensive feature expression of each node through the attention mechanism. These feature expressions capture the importance of the node in the network, including the intensity and pattern of its interaction with adjacent nodes. In this way, each node generates a feature vector that reflects the structure and functional status of the node in the entire network.

[0128] After completing the vulnerability assessment of the device nodes, the key asset verification process can be executed. For the key assets entered by the system engineer, regardless of whether their vulnerability assessment level is classified as "serious", key security maintenance is carried out to protect the security of important assets of the distribution network and prevent major losses to the system. In a resource-constrained environment, targeted security maintenance strategies should be adopted for nodes with different vulnerability levels. For low-vulnerability nodes, they can be monitored and maintained through regular inspections to prevent potential problems from expanding. For medium-vulnerability nodes, it is recommended to repair known vulnerabilities in a timely manner to reduce security risks. High-vulnerability nodes should be treated as a priority to reduce or eliminate potential threats as soon as possible. For nodes with severe vulnerabilities, emergency maintenance must be carried out, and sufficient security measures must be taken to prevent them from causing serious impacts on the entire system, and repair measures should be implemented as soon as possible. This hierarchical maintenance strategy can maximize overall security under limited resources. This process not only realizes the global hierarchical security maintenance management of system equipment, but also effectively guarantees the security of the network.

[0129] The embodiments of this specific implementation method are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Therefore, all equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A distribution network communication network vulnerability assessment method based on graph neural network, characterized in that: include: Acquire the operation data of the distribution network communication network, the operation data including node device attribute data and link attribute data; Based on the operation data of the distribution network communication network, the vulnerability of each node is graded and labeled to form a training data set; Establish a distribution network communication network vulnerability assessment model; The distribution network communication network vulnerability assessment model is trained using the operating data in the training dataset as input and the vulnerability level labels as supervision; Obtain the current distribution network communication network operation data; The distribution network communication network vulnerability assessment model is used to grade the vulnerability of the current distribution network communication network nodes.

2. According to a method for evaluating the vulnerability of a distribution network communication network based on a graph neural network according to claim 1, it is characterized in that: The graded labeling of each node vulnerability includes: Dividing a number of nodes under the distribution network into a first set and a second set; Obtain the vulnerability level label of each node in the first set evaluated by experts; A vulnerability level label is assigned to each node in the second set according to the centrality index of each node in the second set.

3. A distribution network communication network vulnerability assessment method based on graph neural network according to claim 2, characterized in that: According to the centrality index of each node in the second set, each node in the second set is assigned a vulnerability level label: According to the weights of the centrality indicators of the nodes in the second set, a Shannon entropy weighted combination centrality indicator is established, wherein the centrality indicators of the nodes include degree centrality, betweenness centrality, node centrality and node reliability; Calculate the centrality index of the nodes in the second set based on the Shannon entropy weighted combination centrality index; The vulnerability levels of the nodes in the second set are marked according to the centrality indexes of the nodes in the second set.

4. A distribution network communication network vulnerability assessment method based on graph neural network according to claim 3, characterized in that: The establishment of the Shannon entropy weighted combination centrality index includes: Get the centrality index parameters of the node, the centrality index parameters of the node include degree centrality C D , betweenness centrality C B , node centrality C P and node reliability C R ; Determine degree centrality, betweenness centrality, node centrality and entropy measure E of node reliability D 、E P 、E B 、E R ; Determine the weight w of each centrality index under the node i ; Obtain the Shannon entropy weighted combination centrality index: C=w1C D +w2C P +w3C B +w4C R 。 5. The method for evaluating the vulnerability of a distribution network communication network based on a graph neural network according to claim 1 is characterized in that: The distribution network communication network vulnerability assessment model has a graph attention neural network extraction and decision tree ensemble learning mechanism.

6. A distribution network communication network vulnerability assessment method based on graph neural network according to claim 5, characterized in that: The establishment of a distribution network communication network vulnerability assessment model comprises: Establish a graph neural network model for extracting distribution network communication network features; Determine the objective function; Based on the objective function, a decision tree is constructed layer by layer.

7. A distribution network communication network vulnerability assessment method based on graph neural network according to claim 6, characterized in that: The method of using the distribution network communication network vulnerability assessment model to classify the vulnerability of the current distribution network communication network nodes includes: Extracting the communication network feature vector of the distribution network; Based on the feature vector, the vulnerability level of the node is calculated using a decision tree.

8. A distribution network communication network vulnerability assessment method based on graph neural network according to claim 7, characterized in that: The extracting of the distribution network communication network feature vector comprises: Extracting node feature vectors of each node; Extracting link feature vectors of each link; Establish node weight matrix W v and the edge weight matrix W e ; According to the node weight matrix W v and the edge weight matrix W e Perform linear transformation on node feature vectors and link feature vectors respectively; Calculate the attention coefficient of each node; According to the attention coefficient of each node, the feature information of the node neighbors is aggregated.

Images