Electric power system node importance evaluation method based on graph neural network

Abstract

The invention discloses an electric power system node importance evaluation method based on a graph neural network, and the method comprises the following steps: 1), constructing an electric power system topological graph, abstracting an electric power system into a weighted graph G = (V, E), enabling a node V to represent electric power equipment, enabling an edge E to represent an electric power transmission path, and enabling the node and the edge to have electrical attributes; 2) aggregating electrical attributes and topological information of nodes through a graph convolutional network (GCN), and generating a node embedding vector; 3) optimizing node embedding representation by adopting self-supervised contrast learning, and calculating the similarity between the nodes; 4) updating node attributes and topological structures in real time according to dynamic power system monitoring data, and adjusting node importance scores; and 5) selecting the node with the highest importance score through a greedy algorithm, and optimizing the resource allocation of limited monitoring equipment.The method has the beneficial effects that powerful decision support is provided for scheduling, fault diagnosis and emergency response of the power system, and the monitoring efficiency and the overall safety of the power system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring and optimization, and in particular to a method for evaluating the importance of power system nodes based on graph neural network. Background Art

[0002] As an important part of national infrastructure, the stability and reliability of the power system directly affect the normal operation of society and economic development. With the transformation of the global energy structure and the widespread application of renewable energy, the scale and complexity of the power system are increasing. The traditional power system is mainly composed of power generation, transmission and distribution, and the transmission and distribution of electricity rely on fixed power transmission lines and equipment. However, with the access of smart grids and large-scale new energy, the topology of the power system has become more complex, and the dependencies between nodes and the dynamic changes of power flow have also become more complex. How to improve the stability and security of the power system has become a core issue that needs to be urgently solved in the power industry.

[0003] Most of the existing power system monitoring methods rely on traditional physical quantities and topological structures, such as voltage, load, node degree, etc. These methods have certain application effects in simple power grid structures, but when faced with large-scale, complex and dynamically changing power grids, their evaluation results often cannot accurately reflect the criticality of the nodes. Therefore, the applicability of traditional methods has certain limitations, especially in high-dimensional complex power systems. How to effectively evaluate the importance of each node in the power system is still a problem that needs to be solved urgently.

[0004] The node importance assessment of the power system is the basis for ensuring the stable operation of the power system. By accurately assessing the importance of nodes, resource allocation can be optimized in a timely manner when a fault or disaster occurs, the spread of faults can be avoided, and the safety and reliability of the system can be guaranteed. In recent years, with the development of artificial intelligence technologies such as machine learning and deep learning, data-driven node importance assessment methods have gradually become a research hotspot. For example, convolutional neural networks (CNNs) and recurrent neural networks (RNNs) in deep learning have been applied to power system analysis, but these methods usually require a large amount of labeled data, and the acquisition of labeled data in power systems is often very difficult, which limits the effectiveness of these methods in practical applications.

[0005] At the same time, the introduction of complex network theory provides new ideas for evaluating the importance of nodes in power systems. Researchers use indicators such as network structure entropy and betweenness centrality to evaluate the importance of nodes, thereby gaining a more comprehensive understanding of the stability and fault propagation of power systems. However, these methods still face certain challenges when dealing with power system topologies with high-dimensional characteristics, especially when faced with the dynamics and multidimensionality of power systems, which traditional evaluation methods find difficult to handle efficiently.

[0006] To address the deficiencies in the existing technologies and leverage the technical advantages of graph neural networks, a method for evaluating the importance of power system nodes based on graph neural networks is proposed. Graph neural networks can effectively process the topological structure and node attributes in the power system and capture the potential relationships between nodes through graph embedding techniques. This method can not only improve the accuracy of node importance evaluation but also adapt to the dynamic changes in the power system to achieve real-time monitoring and fault prediction. In addition, by using self-supervised learning and contrastive learning strategies, it is possible to effectively evaluate the importance of nodes even in the absence of labeled data, overcoming the limitations of data dependence in traditional methods.

[0007] Overall, the current methods for evaluating the importance of power system nodes still face various challenges, including: the accuracy issue when dealing with complex network topologies, the model training issue when data lacks annotations, and the issue of how to dynamically evaluate the importance of nodes under different power system states. The method for evaluating the importance of nodes based on graph neural networks is expected to solve these problems and provide more scientific and effective technical support for the stable operation and resource optimization of the power system. Summary of the Invention

[0008] In view of the above technical problems to be solved, the present invention provides a method for evaluating the importance of power system nodes based on graph neural networks.

[0009] To solve the above technical problems, the technical solution proposed by the present invention is: a method for evaluating the importance of power system nodes based on graph neural networks, comprising the following steps: 1) Construct a power system topology graph, abstract the power system as a weighted graph G=(V, E), where the nodes V represent power equipment, the edges E represent power transmission paths, and both the nodes and edges have electrical attributes; 2) Aggregate the electrical attributes and topological information of the nodes through a graph convolutional network (GCN) to generate node embedding vectors; 3) Optimize the node embedding representation using self-supervised contrastive learning, calculate the similarity between nodes, and calculate the importance score of the nodes through the following formula: where ei and ej are the embedding representations of nodes vi and vj; sim(ei, ej) is the cosine similarity; |V| is the total number of nodes; 4) Update the node attributes and topological structure in real time according to the dynamic power system monitoring data, and adjust the node importance score; 5) Select the nodes with the highest importance score through a greedy algorithm to optimize the resource allocation of limited monitoring devices.

[0010] As a further improvement of the above technical solution: Preferably, the construction of the power system topology diagram includes the following steps: S11, generating a set V of power equipment nodes and a set E of connection edges between devices by obtaining physical connection information in the power system; S12, assigning electrical attributes to each node according to the electrical characteristics of the power equipment (such as voltage, load, power factor, etc.), and assigning weights to each edge in combination with the power flow capacity; S13, generating a weighted topology diagram, where the weight of each edge reflects the power transmission capacity or impedance, and the node attributes affect the aggregation calculation in the graph convolution process.

[0011] Preferably, the training of the graph convolutional network (GCN) includes: S21, initializing the feature representation of the nodes through the adjacency matrix and the node feature matrix; S22, aggregating and updating the node features using the graph convolutional layer, and calculating according to the following formula: where \(h_i^{(k)}\) represents the representation of node \(v_i\) at the \(k\)-th layer, \(N(i)\) represents the neighbor nodes of node \(i\), and \(\sigma\) is the activation function.

[0012] Preferably, the self-supervised learning and contrastive learning are realized through the following steps: S31, adopting the loss function of graph contrastive learning, and optimizing the embedding representation of the nodes by comparing the differences in the embedding vectors of similar nodes and dissimilar nodes: where \(P\) is the positive sample pair, \(N\) is the negative sample pair, \(sim(·,·)\) represents the cosine similarity, and \(T\) is the temperature parameter.

[0013] Preferably, the node importance is calculated by the following formula: where \(e_i\) and \(e_j\) are the embedding representations of nodes \(v_i\) and \(v_j\), \(sim(e_i, e_j)\) is the similarity of the node embedding representations, and \(|V|\) is the total number of nodes.

[0014] Preferably, the resource allocation optimization of the greedy algorithm includes the following steps: S1. Sort all nodes in descending order according to the dynamically updated node importance scores; S2. Select the highest-scoring uncovered nodes from the sorted list in turn and allocate monitoring devices; S3. Check whether the neighbor nodes of the selected nodes are covered by monitoring devices. If not, mark them as secondary monitoring areas; S4. Repeat steps S2 - S3 until all monitoring devices are allocated or the preset coverage threshold is reached; where the allocation priority of the monitoring devices satisfies the following conditions: The node importance score calculated for claim 1; For the node Topological connectivity; Is the weight coefficient, used to balance the influence of electrical properties and topological structure (default value is 0.5).

[0015] Preferably, the method is applicable to dynamic power system monitoring, updates node attributes and topological structure in combination with real-time data, generates a dynamic importance score, and adjusts the monitoring resources of nodes in real time according to the changes in the power system state.

[0016] Preferably, the method can evaluate the node importance in real time according to the changes in the load, equipment status, etc. of the power system, and adjust the importance ranking through the updated topological structure to ensure the rapid recovery of the power system in case of a fault.

[0017] Preferably, the real-time update of the dynamic power system monitoring data includes: Collect real-time data of node voltage, load, and power through the SCADA system to update node attributes; When a topological change is detected, reconstruct the adjacency matrix A and input it into the graph convolutional network; Adjust the node importance score according to the following rules: For the node Power change amount; Is the maximum allowable power deviation of the system; Is the historical data weight coefficient (default value is 0.7).

[0018] Preferably, the method supports regional management of the power system, divides the power grid into multiple sub-regions, independently evaluates the importance of nodes within each region, and finally aggregates the evaluation results of each region for global optimization.

[0019] Preferably, it further includes a fault diagnosis module, and the specific implementation is as follows: When the node importance score drops suddenly by more than the threshold within the time window , trigger a fault warning: When locating the fault source, first check the top nodes with the highest importance score ; Combined with the similarity clustering results of node embedding vectors, distinguish common cause faults from independent faults. Compared with the prior art, the power system optimization method based on topological compression provided by the present invention has the following beneficial effects: Through the embedding learning of the graph neural network, the present invention can more comprehensively consider the topological structure and electrical characteristics of the nodes in the power system, and realize a more accurate evaluation of node importance; Combining the self-supervised learning and contrast learning methods, the present invention can effectively utilize unlabeled data for training, avoiding the dependence on a large amount of labeled data in traditional methods; The present invention can dynamically evaluate the importance of the nodes in the power system, adapt to the dynamic changes such as load fluctuations and equipment start-stop in the power system, and improve the robustness of the system; The present invention provides an efficient monitoring resource allocation scheme. Through the importance score ranking, it ensures that key nodes are fully monitored, reducing the usage cost of monitoring equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is an example flowchart of the method of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following further describes the present invention in detail with reference to the drawings.

[0022] The method for evaluating the importance of power system nodes based on a graph neural network of the present invention adopts a graph neural network model and a self-supervised learning method to evaluate the importance of nodes in the power system, so as to optimize the resource allocation and monitoring strategy of the power system. The following will detail the implementation steps of the present invention.

[0023] 1. Construction of the power system topology graph The topology graph of the power system is the basis of the present invention. In actual implementation, first, it is necessary to obtain the real-time data of the system and the physical connection relationship of power equipment through the SCADA (Supervisory Control And Data Acquisition) system or other monitoring systems in the power system. Then, abstract these information into a graph structure, where the nodes of the graph represent the equipment in the power system (such as generators, substations, distribution lines, etc.), and the edges represent the power transmission paths between the equipment.

[0024] When constructing the topology graph, each node and edge has its corresponding attributes. The attributes of the nodes include electrical characteristics such as voltage, load, and power, and the attributes of the edges include power transmission capabilities such as transmission capacity and impedance. All these attributes will be used as the basis for subsequent calculation of node importance.

[0025] 2. Graph Convolutional Network (GCN) and Self-Supervised Learning In the implementation of the present invention, a graph convolutional network (GCN) is used to extract the embedded representation of nodes from the power system topology graph. The embedded representation of each node is determined by its own electrical properties and the topological relationship with neighboring nodes. Through graph convolution operations, GCN can effectively aggregate the information of neighbors around nodes and update the representation of each node through multiple layers of convolution.

[0026] To optimize the node embedding, the present invention adopts a self-supervised learning method. Specifically, using the existing structured data (such as voltage, load, etc.) and system state information in the power system, the embedded representation of nodes is optimized through a contrastive learning strategy. Positive sample pairs come from similar nodes, while negative sample pairs come from dissimilar nodes. By maximizing the similarity of positive sample pairs and minimizing the similarity of negative sample pairs, the trained model can more accurately evaluate the importance of nodes.

[0027] 3. Node Importance Evaluation Once the training of node embedding is completed, the next step is to calculate its importance score based on the embedded representation of the node. The importance of a node can be obtained by calculating the similarity between node embedding vectors. For example, the cosine similarity or Euclidean distance can be used to measure the similarity between nodes. The higher the similarity of a node to all other nodes in the system, the higher the importance score of the node.

[0028] To further improve the accuracy and reliability of the evaluation, the present invention also introduces an optimization module. This module uses a greedy algorithm to select the most important nodes from multiple possible sets of monitoring nodes, thus achieving an optimal configuration under limited monitoring device resources.

[0029] 4. Dynamic Node Importance Evaluation The power system is dynamically changing. Therefore, the importance score of nodes also needs to be adjusted as the power system changes. For this purpose, the present invention adopts a real-time data update mechanism, combined with dynamic monitoring data, to continuously adjust the attributes and topological structure of nodes. Especially in the case of load changes, equipment start-stop, and faults in the power system, the importance score of nodes will be recalculated and updated according to the new system state information.

[0030] 5. Node Sorting and Resource Allocation Based on the importance score of nodes, all nodes in the power system can be sorted. Nodes with higher importance are usually key nodes in the system. For these nodes, monitoring devices are preferentially allocated for real-time monitoring. In the case of limited resources, optimization algorithms (such as greedy algorithms) can be used to select the most important nodes for monitoring, thereby improving the stability and security of the system.

[0031] 6. Fault Diagnosis and Recovery In practical applications, when a system fails, the importance score of nodes can help quickly identify the source of the failure. For example, nodes with a relatively high importance score in the system (such as key substations or load nodes) may have a greater impact on the stability of the system once they fail. Therefore, by combining node importance assessment, potential failure sources can be detected earlier and restored through optimized scheduling strategies.

[0032] 7. Real-time Monitoring and Decision Support The implementation of the present invention can also provide support for the real-time monitoring and decision-making of power systems. By combining node importance assessment with the scheduling optimization strategy of power systems, the operation mode of the power grid can be quickly adjusted during dynamic load changes and emergencies. Especially in emergency situations, the importance of critical nodes can be evaluated in real time and quick decisions can be made, which can effectively improve the robustness and response speed of power systems.

[0033] 8. Implementation Adaptable to Different Power Systems According to the scale and complexity of power systems, the present invention provides flexible implementation methods. For small power grids, relatively simple models can be used for node evaluation and resource allocation; while for large-scale power grids, distributed computing and hierarchical optimization methods are required to locally optimize different regions of the power grid and finally achieve the resource scheduling and fault recovery of the entire network.

[0034] Experimental Data Demonstration: Based on multiple publicly available power system datasets (IEEE 300-bus system) for benchmark comparison, the performance of different methods is evaluated. Through the evaluation of these standard power systems, we can obtain the metrics of different methods in node importance assessment: Table 1 Performance Comparison of the Proposed GNN Scheme with Other Schemes Evaluation method Node importance accuracy (%) Computation time (minutes) Resource allocation savings (%) Ability to adapt to dynamic changes (response time, minutes) Degree Centrality 57% 2 0% 5 Betweenness Centrality 61% 3 0% 5 GNN (Graph Neural Network) 85% 5 17.5% 2 Deep learning method based on CNN 72% 8 9% 4 LSTM 77% 9 6% 4 Compared with the above traditional methods and other deep learning methods, the method based on graph neural network shows significant advantages in the following aspects: Node Importance Accuracy: The method based on GNN effectively integrates the topological structure and electrical attributes of power systems through graph convolutional networks, improving the accuracy by 28%-30% compared with traditional degree centrality and betweenness centrality methods, and by 13%-14% compared with deep learning methods based on CNN.

[0035] Computation Time: The computation time of the method based on GNN is 5 minutes, which is better than the 9-minute computation time of the method based on LSTM. Compared with traditional methods, the computation time of GCN is shorter, providing a faster evaluation speed.

[0036] Resource allocation savings: The GNN method optimizes resource allocation through accurate node importance scoring and a greedy algorithm, which can save 17.5% of monitoring resources. In contrast, the CNN-based method saves 9%, and Degree Centrality and Betweenness Centrality do not optimize resource allocation, with a savings of 0%.

[0037] Ability to adapt to dynamic changes: The GNN-based method can respond to load fluctuations and equipment start / stop in the power system within 2 minutes and adjust the node importance score in real time. In contrast, traditional methods take 5 minutes, and deep learning methods based on CNN and LSTM take 4 minutes.

[0038] To support efficient computing for large-scale power system datasets (such as IEEE 118-node and IEEE 300-node systems), the recommended server configuration is as follows: Table 2 Server Configuration Details This configuration can support the training and evaluation of graph neural networks for large-scale power grid data, ensuring real-time adjustment of node importance scores and optimization of resource allocation. GPU acceleration can speed up the training of graph convolutional layers to adapt to real-time load fluctuations and dynamic changes.

[0039] Compared with traditional node importance evaluation methods, the present invention has significant advantages. First, the graph neural network (GCN) can better capture the complex relationships between nodes in the power system and provide more accurate node embedding representations. Second, the self-supervised learning-based method can reduce the dependence on a large amount of labeled data and improve the flexibility and adaptability of evaluation. Finally, the combination of dynamic evaluation of node importance and real-time monitoring enables the method to handle changes and emergencies in the power system, enhancing the security and stability of the power system.

[0040] This specific embodiment is only an explanation of the present invention and is not a limitation thereof. Those skilled in the art can make modifications without creative contributions to this embodiment after reading this specification, but as long as they are within the scope of the claims of the present invention, they are protected by the patent law.

Claims

1. A method for evaluating the importance of power system nodes based on graph neural networks, characterized in that Including the following steps: 1) Construct a power system topology graph, abstract the power system as a weighted graph G=(V,E), where the nodes V represent power equipment, the edges E represent power transmission paths, and both the nodes and edges have electrical attributes; 2) Aggregate the electrical attributes and topological information of the nodes through a graph convolutional network (GCN) to generate node embedding vectors; 3) Use self-supervised contrastive learning to optimize the node embedding representation, calculate the similarity between nodes, and calculate the importance score of the nodes through the following formula: where ei and ej are the embedding representations of nodes vi and vj; sim(ei,ej) is the cosine similarity; |V| is the total number of nodes; 4) Update the node attributes and topological structure in real time according to the dynamic power system monitoring data, and adjust the node importance score; 5) Select the nodes with the highest importance score through a greedy algorithm to optimize the resource allocation of limited monitoring devices.

2. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, wherein The construction of the power system topology graph includes: S11, obtaining the physical connection information of power equipment to form a node set V and an edge set E; S12, assigning attribute values to each node and edge, including voltage, load, power, impedance, etc.; S13, generating a weighted topology graph, where the weight of the edge reflects the power transmission capacity or impedance.

3. The method for evaluating the importance of power system nodes based on graph neural network according to claim 2, characterized in that, The training process of the graph convolutional network includes: S21, initializing the node representation through the adjacency matrix and the feature matrix; S22, updating the node representation using the graph convolutional layer according to the following formula: where N(i) is the neighbor of node i, di represents the degree of node i, W(k),b(k) are the weight and bias parameters, and σ is the activation function.

4. The power system node importance evaluation method based on a graph neural network according to claim 3, wherein Self-supervised contrastive learning uses the following contrastive loss function to optimize the node embedding: where P is the positive sample pair, N is the negative sample pair, sim(·,·) is the cosine similarity, and T is the temperature parameter; The importance of the node is calculated through the following formula: where ei,ej are the embedding representations of nodes i,j, and |V| is the total number of nodes.

5. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, characterized in that, The resource allocation optimization of the greedy algorithm includes the following steps: S1. Sort all nodes in descending order according to the dynamically updated node importance score; S2. Select the highest-scoring nodes that have not been covered in turn from the sorted list and allocate monitoring devices; S3. Check whether the neighbor nodes of the selected nodes have been covered by monitoring devices. If not, mark them as secondary monitoring areas; S4. Repeat steps S2 - S3 until all monitoring devices are allocated or the preset coverage threshold is reached; where the allocation priority of the monitoring devices satisfies the following conditions: Node importance score calculated for claim 1; For a node of the topological connectivity; is the weight coefficient, which is used to balance the influence of electrical properties and topological structures (the default value is 0.5).

6. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, characterized in that Suitable for dynamic power system monitoring, updating node attributes and topological structure in combination with real-time monitoring data to generate dynamic importance scores.

7. A method for evaluating the importance of power system nodes based on graph neural network according to claim 1, characterized in that Suitable for power grid subgraphs containing high-load nodes and key substations, used to prioritize the monitoring and fault prediction of important nodes.

8. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, characterized in that, The real-time update of the dynamic power system monitoring data includes: Collecting the real-time data of node voltage, load, and power through the SCADA system to update the node attributes; When a topological change is detected, reconstruct the adjacency matrix A and input it into the graph convolutional network; Adjust the node importance score according to the following rules: is the power change amount of the node ; is the maximum allowable power deviation of the system; It is the weight coefficient of historical data (the default value is 0.7).

9. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, wherein The power system includes multiple regions. The node importance of the power grid subgraphs within each region is evaluated by a local optimization method, and finally the evaluation results of each region are summarized for global resource optimization.

10. The method for evaluating the importance of power system nodes based on graph neural network according to claim 1, characterized in that, It also includes a fault diagnosis module, and the specific implementation is as follows: When the node importance score drops suddenly by more than the threshold within the time window a fault warning is triggered: When locating the fault source, first check the top nodes with the highest importance score ; Combined with the similarity clustering results of node embedding vectors, common-cause faults and independent faults are distinguished.

Images