A graph neural network-based feeder fast tracing method

Abstract

The application provides a kind of based on graph neural network's feeder fast tracing method. The method realizes the fast tracing and visual display of power supply path through automated means, not only improves the accuracy and timeliness of tracing, but also greatly reduces the workload of dispatch personnel, improves the reliability of power grid safe operation. The method includes power grid topology modeling, real-time operation data reading, GNN modeling, A* algorithm tracing and visual display module.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power systems, in particular to a feeder fast tracing method based on a graph neural network. BACKGROUND

[0002] With the continuous development of power systems and the expansion of power grid scale, the complexity of power system operation is also increasing. In order to alleviate the problem of short-circuit current exceeding the standard, power grid layering and zoning has become a trend. However, the frequency of 10kV system crossing 500kV substations for loop closing and opening operations is increasing day by day. Such operations involve upper-level dispatching permission and large-scale electromagnetic loop networks, which challenge the safety level of the distribution network itself. Due to the large voltage level and dispatching level, the sensitivity of county-level distribution dispatching to such operations is low, which may cause operation risks and affect the safety and stability of power grid operation. Therefore, how to quickly and accurately analyze the power supply of the 10kV feeder involved in loop closing and opening has become a technical problem to be solved.

[0003] At the same time, the number of various types of power protection work on the distribution network side undertaken by the current power company is growing, showing the characteristics of "many points and wide range". According to relevant regulations, the protection range of users needs to cover high-voltage equipment directly supplied from 10kV distribution feeders to 220kV substations. The existing protection range demarcation is usually traced by dispatchers according to the 10kV protection user list one by one, and a protection and supply diagram is drawn to realize accurate monitoring of the protection equipment. This method is time-consuming, tedious and prone to errors, which affects the efficiency of dispatchers' other dispatching work. In addition, with the large-scale access of various types of new energy to the power grid, the situation of power reverse feeding appears in some areas. The tracing method considering only power flow is no longer applicable to the new power system, and traditional graph search algorithms such as depth-first search (DFS) and breadth-first search (BFS) have problems of insufficient efficiency and flexibility when dealing with complex power grid structure and dynamic load data. With the increasing complexity and intelligentization of power systems, there is an urgent need for a method that can intelligently and dynamically obtain the power supply path.

[0004] In domestic and foreign countries, real-time topology of power grid has been widely used in energy management system (EMS) for power grid equipment live state analysis, accident tripping research and judgment and other application fields. However, the technology based on regional EMS system to realize visual display of power supply path and assist dispatchers in power supply source tracing and power supply safety analysis is also in urgent need of development. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a feeder fast tracing method based on a graph neural network to solve the problems in the prior art, improve the efficiency and accuracy of power protection work, and reduce the work burden of dispatchers.

[0006] To achieve the above object, the application adopts the following technical scheme: A feeder fast tracing method based on a graph neural network, comprising the following steps:

[0007] Step 1: modeling of power grid topology available for tracing analysis;

[0008] Step 2: real-time operation data reading;

[0009] Step 3: modeling of a graph neural network GNN;

[0010] Step 4: A*(A-star) algorithm for power supply path tracing;

[0011] Step 5: visual display.

[0012] In a preferred embodiment, the step 1 specifically comprises: obtaining information of various components of a current power grid from a power grid device model library, the various components of the current power grid including substations, transformers, busbars, circuit breakers, disconnectors, cable lines and overhead lines, standardizing and classifying the above information to form a standardized device model database;

[0013] On the basis of the establishment of the device model library, the connection relationship between devices is inputted, the connection relationship between devices including the connection between a transformer and a busbar, the connection between a busbar and a circuit breaker, and the connection between a circuit breaker and a line; meanwhile, the system is set to periodically update the above device library and topological relationship;

[0014] After obtaining the device model and topological connection relationship data, these data need to be comprehensively processed and applied to support fast tracing analysis of the power supply path; based on the integrated data, a power grid topological model is established; graph theory algorithm is used to model the devices and connection relationship as a graph structure, wherein the nodes represent the devices and the edges represent the connection relationship; a standardized data interface is designed to interact with the power user electricity information collection system and the EMS system, to ensure real-time updating of the device model and topological data; the power grid topological model is integrated into the power dispatching management platform to provide a unified data access and management interface, to support power supply path tracing and fault analysis application.

[0015] In a preferred embodiment, the step 2 specifically comprises: obtaining operation mode data and device power flow data of power grid devices from the EMS system; the operation mode data and device power flow data include the current operation state, voltage level and current direction of the devices; by combining the device model with real-time data, a topological analysis model data of the power grid is formed; the topological analysis model is the basis for fast tracing of the power supply path, to ensure that the real-time operation state of the power grid can be reflected;

[0016] The monitoring data of the operating state and current size of the switch are updated to the topology relationship formed in step one, and the updated topology model can truly reflect the current power grid operation mode and ensure the accuracy of the subsequent tracing algorithm.

[0017] In a preferred embodiment, the step 3 specifically comprises:

[0018] Step 31: initial modeling:

[0019] Initially, the power grid is modeled as a graph G:

[0020] G=(V,E)

[0021] Wherein, V is a set of nodes, including all substation, bus and feeder node information; E is a set of edges, representing the connection relationship between nodes; the initial feature vector h of each node v (0) Contains voltage level, connection information and switch state;

[0022] Step 32: topology structure change processing:

[0023] Topology changes mainly include the following cases:

[0024] Addition and removal of nodes: the addition of nodes may be a new substation or feeder, and the removal of nodes may be the shutdown or removal of some equipment; addition and deletion of edges: the addition of edges indicates the connection line between the newly added nodes, and the deletion of edges indicates the disconnection or removal of the existing connection line;

[0025] The graph structure of GNN is dynamically updated, including the following aspects:

[0026] (1) Adjacency matrix and degree matrix update

[0027] The adjacency matrix A represents the connection relationship between nodes in the graph, which is an N×N matrix, N is the number of nodes; if a node or edge is added or removed in the graph, the adjacency matrix needs to be increased or decreased in rows and columns, and the non-zero elements in the matrix need to be updated; if a node is added, a new row and column are added to the adjacency matrix to represent the connection relationship of the node; if a node is removed, the corresponding row and column are deleted from the adjacency matrix; if an edge is added or deleted, the corresponding adjacency matrix element is modified to reflect the added or deleted connection relationship;

[0028] The degree matrix D is a diagonal matrix, whose elements are the degrees of the nodes; after updating the adjacency matrix, the degree matrix also needs to be updated accordingly: adding an edge: increasing the degree of the related node; deleting an edge: reducing the degree of the related node;

[0029] (2) Edge weight and feature update

[0030] In many cases, the weight and feature of the edge can also change, which requires dynamic update:

[0031] Edge weight update: In the adjacency matrix, the weight of the edge reflects the characteristics of the line in the power grid; if the switch state changes or the line characteristics change, the corresponding element in the adjacency matrix needs to be updated;

[0032] Feature matrix update: the edge feature matrix stores additional information of the edge, which needs to be updated to reflect the current system state when the state changes;

[0033] Step 33: training the graph neural network GNN:

[0034] The basic operation of the graph convolution network GCN is to aggregate and update the features of each node with its neighbors; in a dynamic environment, the changes of the adjacency matrix and node features require us to flexibly adjust the graph convolution operation;

[0035] The standard graph convolution operation is defined as:

[0036]

[0037] where H (l) is the node feature matrix of the l-th layer, W (l) is the weight matrix of the l-th layer, σ is the activation function, is the adjacency matrix plus the self-loop, is the normalized degree matrix.

[0038] Then the GCN model is trained, the initial feature matrix H (0) is input, the high-dimensional feature representation of the node is extracted through multiple layers of graph convolution, and finally the high-dimensional feature representation of the node H (L) is obtained, which is used for the design of the heuristic function in the path search.

[0039] In a preferred embodiment, the step 4 specifically comprises:

[0040] Step 41, initialization: put the starting node into the open list and set its actual cost and heuristic cost; the actual cost g of the starting node is 0, and the heuristic cost h is calculated by the heuristic function;

[0041] Step 42, node selection: select the node with the minimum total cost f=g+h from the open list as the current processing node;

[0042] Step 43, check the target node: judge whether the current node is the target node; if yes, the search is ended and the path is backtracked; if not, continue to expand the node;

[0043] Step 44, extension node: obtain all neighbor nodes of the current node, and calculate their actual cost g and heuristic cost h;

[0044] Step 45, open list and closed list update: add the neighbor nodes meeting the condition to the open list, and update the parent node and cost of the node if it already exists and the new path is better; after the current node is processed, it is moved to the closed list, indicating that it has been processed;

[0045] Step 46, path backtracking: if the target node is found, backtrack from the target node to the starting point to build a complete power supply path;

[0046] Step 47, repeat the process: if the open list is not empty, repeat the above process until the target node is found or the open list is empty.

[0047] In a preferred embodiment, the step 5 is specifically: first, study the power supply path analysis result based on the power protection user, realize the matching with the main device standard model, identify the same device in each power supply path, and arrange in groups according to the transformer substation to which the device belongs and the voltage level of the device, complete the power supply path preprocessing;

[0048] Secondly, according to the main device model information, including device type, transformer winding type, device connection relationship and the like, the information such as the used graph element and the size of the graph element of the main device is identified, the graph element layout is carried out according to the device grouping, and the graph element overlap verification is carried out, and the overlapping part is readjusted;

[0049] Finally, the power supply path graph of the power protection user is generated according to the adjusted device graph element position and graph element connection relationship, and is displayed based on the iES700 man-machine interface, and table data export is supported.

[0050] Compared with the prior art, the present application has the following beneficial effects: the accuracy and timeliness of the feeder tracing are improved, the workload of the dispatchers is greatly reduced, and the safety and reliability of the distribution network operation are improved. BRIEF DESCRIPTION OF DRAWINGS

[0051] Fig. 1 is the system overall structure schematic diagram of the present application;

[0052] Fig. 2 is the schematic diagram of the A* heuristic search algorithm based on the graph neural network;

[0053] Fig. 3 is the power supply path visualization display method based on the main device standard model. DETAILED DESCRIPTION

[0054] The present application will be further described below in combination with the drawings and embodiments.

[0055] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0056] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application; as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, it will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components and / or combinations thereof.

[0057] A graph neural network-based feeder fast tracing method, referring to Figs. 1-3 , mainly includes the following steps:

[0058] Step one: power grid topology modeling available for tracing analysis.

[0059] Obtain information of various components of the current power grid through the power grid equipment model library, including substations, transformers, buses, circuit breakers, disconnectors, cable lines, overhead lines, standardize the above information, and form a standardized equipment model database.

[0060] On the basis of the establishment of the equipment model library, the connection relationship between each device is input, including the connection of transformer and bus, the connection of bus and circuit breaker, the connection of circuit breaker and line, etc. At the same time, the system is set to update the above device library and topology relationship regularly to ensure that the device can be updated and replaced in the first time after the power grid equipment changes.

[0061] After obtaining the equipment model and topology connection relationship data, these data need to be comprehensively processed and applied to support the fast tracing analysis of power supply path. Based on the integrated data, the power grid topology model is established. Using graph theory algorithm, the equipment and connection relationship are modeled as graph structure, where the node represents the equipment and the edge represents the connection relationship. Design a standardized data interface to interact with the power user electricity information collection system, EMS system and other data to ensure real-time update of equipment model and topology data. The power grid topology model is integrated into the power dispatching management platform to provide a unified data access and management interface, supporting power supply path tracing, fault analysis and other applications.

[0062] Step two: real-time operation data reading.

[0063] The operation mode data and device power flow data of the power grid equipment are obtained from the EMS system. These data include the current operating state of the equipment (such as switch state, load condition), voltage level and current direction, etc. By combining the device model with real-time data, the topological analysis model data of the power grid is formed. The topological analysis model is the basis for fast power supply path tracing, ensuring that the real-time operating state of the power grid can be reflected.

[0064] The monitoring data of the operating state and current size of the switch can be updated to the topological relationship formed in step one. The updated topological model can truly reflect the current operation mode of the power grid, ensuring the accuracy of the subsequent tracing algorithm.

[0065] Step three: graph neural network (GNN) modeling.

[0066] Based on the model formed in steps one and two for tracing analysis, step three uses graph neural network (GNN) for modeling.

[0067] The power grid is essentially a complex network graph, with nodes (substations, buses, and feeders) and edges (connecting lines) forming a multi-level topological structure. GNN can efficiently process graph structure data, capturing the complex relationships and interdependencies between nodes, and has a unique advantage in analyzing the power grid: GNN can directly operate on the graph structure without converting the graph into other forms. GNN can extract information from the local neighbors of nodes through multi-layer graph convolution operations and aggregate features layer by layer to form high-dimensional node representations. This feature learning method can capture the complex relationships and dependencies between nodes. The following is the detailed process of GNN modeling and training:

[0068] 1. Initial modeling:

[0069] Initially, the power grid is modeled as a graph G:

[0070] G = (V, E)

[0071] Where V is the node set, including all substation, bus and feeder node information; E is the edge set, representing the connection relationship between nodes. The initial feature vector h v (0) contains voltage level, connection information and switch state.

[0072] 2. Topological structure change processing:

[0073] In power grid supply path tracing, topology changes are a common and important problem. These changes include the addition and removal of nodes (such as substations and feeders) and the adjustment of connections between nodes (such as the addition or deletion of lines). To effectively handle these changes and maintain the real-time performance and accuracy of graph neural network (GNN) models, a flexible topology change handling method is needed.

[0074] Topology changes mainly include the following cases:

[0075] Addition and removal of nodes: The addition of a node may be a new substation or feeder, and the removal of a node may be the shutdown or removal of some equipment; addition and deletion of edges: the addition of an edge indicates the addition of a connection line between nodes, and the deletion of an edge indicates the disconnection or removal of an existing connection line.

[0076] To cope with dynamic changes in the topology structure, the graph structure of the GNN needs to be dynamically updated, mainly involving the following aspects:

[0077] (1) Adjacency matrix and degree matrix update

[0078] The adjacency matrix A represents the connection relationship between nodes in the graph and is an N × N matrix, where N is the number of nodes. If a node or edge is added or removed from the graph, the adjacency matrix needs to be increased or decreased in rows and columns, and the non-zero elements in the matrix need to be updated; if a node is added, a new row and column are added to the adjacency matrix to represent the connection relationship of the node; if a node is removed, the corresponding row and column are deleted from the adjacency matrix; if an edge is added or deleted, the corresponding adjacency matrix elements are modified to reflect the added or deleted connection relationship.

[0079] The degree matrix D is a diagonal matrix whose elements are the degrees of the nodes (i.e., the number of edges connected to the node). After updating the adjacency matrix, the degree matrix also needs to be updated accordingly: add edges: increase the degree of the relevant nodes; delete edges: reduce the degree of the relevant nodes.

[0080] (2) Edge weight and feature update

[0081] In many cases, the weight and features of the edges (such as switch state, current flow direction, etc.) may also change, and these changes need to be dynamically updated:

[0082] Edge weight update: In the adjacency matrix, the weight of an edge can reflect the characteristics of a line in the power grid (such as transmission capacity or switch state). If the switch state changes or the line characteristics change, the corresponding elements in the adjacency matrix need to be updated.

[0083] Feature matrix update: The edge feature matrix stores additional information about the edges (such as switch state, current direction), and needs to be updated to reflect the current system state when the state changes.

[0084] The topology of the main network and the change of the switch are usually not too frequent, so it is not always necessary to completely retrain the GNN model when it changes. This algorithm adopts a local adjustment strategy, which only updates the nodes and edges directly affected by the topology change, instead of completely recalculating the feature representation of the entire graph, thereby reducing the computational power consumption of retraining the GNN after each change.

[0085] 3. Training the graph neural network (GNN):

[0086] The basic operation of the graph convolutional network (GCN) is to aggregate and update the features of each node with its neighbors. In a dynamic environment, the changes in the adjacency matrix and node features require us to flexibly adjust the graph convolution operation.

[0087] The standard graph convolution operation is defined as:

[0088]

[0089] where H (l) is the node feature matrix of the l-th layer, W (l) is the weight matrix of the l-th layer, σ is the activation function (using the ReLU function), is the adjacency matrix with self-loop added, is the normalized degree matrix.

[0090] Then the GCN model is trained, the initial feature matrix H (0) is input, the high-dimensional feature representation of the node is extracted through multiple layers of graph convolution layer, and finally the high-dimensional feature representation of the node H (L) is obtained, which is used for the design of heuristic function in path search.

[0091] Step four: A*(A-Star) algorithm for path search.

[0092] A* algorithm is a classic heuristic search algorithm, which combines the comprehensiveness of breadth-first search (BFS) and the heuristic idea of greedy algorithm, uses heuristic function to effectively guide the search process, and reduces the possibility of invalid search.

[0093] The core of A* algorithm is to construct a suitable heuristic function to guide the search more efficiently and accurately. The following is the construction description of several functions in the algorithm.

[0094] Total cost function f:

[0095] f(n)=g(n)+h(n)

[0096] In the formula, n represents the current node, g(n) is the actual cost function, and h(n) is the heuristic function.

[0097] The actual cost is the cumulative path cost from the starting node s to the current node n. Since the tracing path is usually unique, the actual cost can be simplified to the number of nodes on the path, i.e.:

[0098] g(n) = num(n)

[0099] num(n) represents the number of nodes on the direct path from the starting node s to the current node n.

[0100] The heuristic function h(n) uses the feature representation extracted by GCN to estimate the cost from the current node to the target node, which can be expressed as:

[0101] h(n) = ||h n -h t ||2

[0102] In the formula, h n h is the GCN eigenvector of node n. t Let be the GCN feature vector of the target node t, and ||·||2 represent the Euclidean distance.

[0103] The specific steps of the A* algorithm used in this method are as follows:

[0104] The first step is initialization: add the starting node (10kV feeder) to the open list and set its actual cost and heuristic cost. The actual cost g of the starting node is 0, and the heuristic cost h is calculated using a heuristic function.

[0105] The second step is node selection: From the open list, select the node with the smallest total cost f = g + h as the current processing node. This step ensures that we always prioritize processing potential shortest paths.

[0106] Step 3: Check the target node: Determine if the current node is the target node (500kV substation). If yes, the search ends and the path is traced back. If not, continue expanding the node list.

[0107] Step 4: Expand nodes: Obtain all neighboring nodes of the current node and calculate their actual cost g and heuristic cost h.

[0108] Step 5: Update Open and Closed Lists: Add neighboring nodes that meet the criteria to the open list. If the node already exists and the new path is better, update its parent node and cost. Once the current node has been processed, move it to the closed list to indicate that processing is complete.

[0109] Step 6: Path backtracking: If the target node is found, backtrack from the target node to the starting point to construct a complete power supply path.

[0110] Step 7: Repeat the process: If the open list is not empty, repeat the above process until the target node is found or the open list is empty (i.e., there is no path to explore).

[0111] The flowchart of the algorithm is as follows: Fig. 2 As shown.

[0112] Step 5: Visualization.

[0113] First, the study studies the matching of power supply path analysis results of power supply users with the standard model of main equipment, identifies the same equipment in each power supply path, and groups and arranges them according to the substation to which the equipment belongs and the voltage level of the equipment, thus completing the power supply path preprocessing.

[0114] Secondly, based on the main equipment model information, including equipment type, transformer winding type, equipment connection relationship, etc., the graphic elements used by the main equipment and the size of the graphic elements are identified. The graphic elements are laid out according to the equipment group, and the graphic element overlap is checked. The overlapping parts are readjusted.

[0115] Finally, based on the adjusted device element positions and element connection relationships, a power supply path diagram for power supply users is generated and displayed on the iES700 human-machine interface, while also supporting tabular data export.

[0116] Specific visualization methods are as follows: Fig. 3 As shown.

Claims

1. A fast feeder source tracing method based on graph neural networks, characterized in that... Includes the following steps: Step 1: Modeling the power grid topology for source tracing analysis; Step 2: Real-time data reading; Step 3: Modeling the Graph Neural Network (GNN); Step 4: Use the A* (A-star) algorithm to trace the power supply path; Step 5: Visual presentation; Step 1 specifically involves: obtaining information on various components of the current power grid through the power grid equipment model library. These components include substations, transformers, busbars, circuit breakers, disconnect switches, cable lines, and overhead lines. The information is then standardized and categorized to form a standardized equipment model database. Based on the establishment of the equipment model database, the connection relationships between various devices are entered, including the connection between transformers and busbars, the connection between busbars and circuit breakers, and the connection between circuit breakers and lines; at the same time, the system is configured to periodically update the above equipment model database and topology relationships. After acquiring the equipment model database and topology connection data, this data needs to be comprehensively processed and applied to support rapid power supply path tracing analysis. Based on the integrated data, a power grid topology model is established. Graph theory algorithms are used to model the equipment and connection relationships as a graph structure, where nodes represent equipment and edges represent connections. Standardized data interfaces are designed to interact with the power user electricity consumption information collection system and EMS system to ensure real-time updates of the equipment model database and topology data. The power grid topology model is integrated into the power dispatch management platform to provide a unified data access and management interface, supporting power supply path tracing and fault analysis applications. Step 4 specifically includes: Step 41, Initialization: Add the starting node to the open list and set its actual cost and heuristic cost; the actual cost g of the starting node is 0, and the heuristic cost h is calculated using a heuristic function; Step 42, Node selection: Select the node with the smallest total cost f=g+h from the open list as the current processing node; Step 43, check the target node: determine if the current node is the target node; if so, end the search and backtrack the path; if not, continue to expand the nodes; Step 44, Expand Nodes: Obtain all neighboring nodes of the current node and calculate their actual cost g and heuristic cost h; Step 45, Open and Closed List Update: Add neighboring nodes that meet the conditions to the open list. If the node already exists and the new path is better, update its parent node and cost. After the current node is processed, move it to the closed list to indicate that it has been processed. Step 46, Path backtracking: If the target node is found, backtrack from the target node to the starting point to construct a complete power supply path; Step 47, Repeat the process: If the open list is not empty, repeat the above process until the target node is found or the open list is empty.

2. The fast feeder source tracing method based on graph neural networks according to claim 1, characterized in that, Step 2 specifically includes: obtaining the operating mode data and power flow data of the power grid equipment from the EMS system; the operating mode data and power flow data include the current operating status of the equipment, voltage level and current direction; by combining the equipment model database with real-time data, the topology analysis model data of the power grid is formed; the topology analysis model is the basis for rapid tracing of power supply paths, ensuring that it can reflect the real-time operating status of the power grid; The monitoring data on the operating status of the switch and the magnitude of the current are updated into the topology relationship formed in step one. Only the updated topology model can truly reflect the current power grid operation mode and ensure the accuracy of subsequent tracing algorithms.

3. The fast feeder source tracing method based on graph neural networks according to claim 1, characterized in that, Step 3 specifically includes: Step 31: Initial Modeling Initially, the power grid is modeled as Figure G: Where V is the set of nodes, including information on all substation, busbar, and feeder nodes; E is the set of edges, representing the connections between nodes; and each node's initial feature vector... Includes voltage level, connection information, and switch status; Step 32: Topology Change Handling: Topological changes mainly include the following situations: Adding and removing nodes: Adding a node may be a new substation or feeder, while removing a node is the shutdown or removal of certain equipment; Adding and deleting edges: Adding an edge represents the connection line between new nodes, while deleting an edge represents disconnecting or removing an existing connection line. Dynamically updating the graph structure of GNNs includes the following aspects: (1) Adjacency matrix and degree matrix update The adjacency matrix A represents the connections between nodes in the graph. It is an N×N matrix where N is the number of nodes. If a node or edge is added or removed from the graph, the adjacency matrix needs to be adjusted accordingly, with rows and columns added or removed, and non-zero elements updated. If a node is added, the adjacency matrix is ​​expanded with a new row and column to represent the connections of that node. If a node is removed, the corresponding row and column are deleted from the adjacency matrix. If an edge is added or removed, the corresponding elements of the adjacency matrix are modified to reflect the added or removed connections. The degree matrix D is a diagonal matrix whose elements are the degree of the nodes. After updating the adjacency matrix, the degree matrix also needs to be updated accordingly: adding an edge increases the degree of the related node; deleting an edge decreases the degree of the related node. (2) Edge weights and feature updates In many cases, the weights and features of edges may also change, and these changes need to be updated dynamically: Edge weight update: In the adjacency matrix, the weight of the edge reflects the characteristics of the lines in the power grid; if the switch state changes or the line characteristics change, the corresponding elements in the adjacency matrix need to be updated. Feature matrix update: The edge feature matrix stores additional information about the edges and needs to be updated when the state changes to reflect the current system state; Step 33: Training the Graph Neural Network (GNN): The basic operation of Graph Convolutional Networks (GCNs) is to aggregate and update the features of each node with the features of its neighbors. In dynamic environments, changes in the adjacency matrix and node features require us to flexibly adjust the graph convolution operation. The standard graph convolution operation is defined as: ,in, It is the first l The node feature matrix of the layer It is the first l The weight matrix of the layer, It is an activation function. It is an adjacency matrix with self-loops. It is a normalized degree matrix; Then, the GCN model is trained, with the initial feature matrix as input. H (0) High-dimensional feature representations of nodes are extracted through multiple graph convolutional layers, ultimately yielding the high-dimensional feature representations of the nodes. H (L) This is used for heuristic function design in path search.

4. The fast feeder source tracing method based on graph neural networks according to claim 1, characterized in that, Step 5 specifically involves: First, studying the power supply path analysis results based on the power supply users, achieving matching with the standard model of the main equipment, identifying the same equipment in each power supply path, and grouping and arranging them according to the substation to which the equipment belongs and the voltage level of the equipment, thus completing the power supply path preprocessing; Secondly, based on the main equipment model information, including equipment type, transformer winding type and equipment connection relationship, the size information of the graphic elements used by the main equipment is identified, the graphic elements are laid out according to the equipment group, and graphic element overlap is checked and the overlapping parts are readjusted. Finally, based on the adjusted device element positions and element connection relationships, a power supply path diagram for power supply users is generated and displayed on the iES700 human-machine interface, while also supporting tabular data export.

Images