Distribution network fault location method and system

Abstract

The present invention relates to the technical field of distribution network fault location, and discloses a distribution network fault location method and system. The method comprises constructing an initial topology graph G0 = (V0, E0) of the distribution network; triggering an incremental update module when a topology change event occurs to obtain an incremental node set and incremental edge set affected by the topology change event, and using this to update the initial topology graph to obtain a target topology graph G1; constructing a graph dataset H = (G1, Xt) based on the target topology graph and multi-channel time series data Xt, inputting the graph dataset into a graph neural network model to obtain a fault propagation probability matrix; identifying suspected fault areas based on the fault propagation probability matrix, synchronously injecting pulse signals into nodes within the suspected fault areas, and collecting node response signals; and locating the distribution network fault location based on the fault propagation probability matrix and node response signals. The method and system of the present invention achieve rapid and accurate fault location in the event of dynamic changes in the distribution network topology.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault location, and in particular to a distribution network fault location method and system. Background Art

[0002] As the power system develops towards intelligence and automation, the distribution network, as the terminal link of the power system, faces increasingly higher requirements for its operational safety and power supply reliability, requiring rapid location of the fault when a fault occurs.

[0003] Most existing distribution network fault location methods are based on fixed topologies. They identify faulty equipment or lines by monitoring single-point electrical quantities or analyzing historical data. These methods ignore the dynamic changes in topology during network operation and are unable to adapt to real-world conditions such as frequent switching operations and the switching of distributed power sources on and off the grid. This can easily lead to positioning errors or misjudgments. Furthermore, while some methods incorporate graph neural networks (GNNs) to model topological information, the input graph structure is static and cannot be updated in real time with operating conditions. This leads to problems such as delayed response and poor adaptability. Summary of the Invention

[0004] To this end, the purpose of the present invention is to overcome the problem of distribution network fault location deviation or misjudgment in the prior art, and to provide a distribution network fault location method and system to quickly and accurately locate the fault location when the distribution network topology structure dynamically changes.

[0005] In a first aspect, to solve the above technical problems, the present invention provides a distribution network fault location method, comprising:

[0006] Construct the initial topology graph G0 = (V0, E0) of the distribution network; V0 represents the node set; E0 represents the edge set; among them, power equipment is mapped as a node; the connection relationship between power equipment is mapped as an edge;

[0007] When a topology change event occurs, the incremental update module is triggered to obtain an incremental node set and an incremental edge set affected by the topology change event, and the initial topology graph G0 is updated according to the incremental node set and the incremental edge set to obtain a target topology graph G1;

[0008] Obtain multi-channel time series data Xt of the node when the fault occurs, construct a graph dataset H=(G1, Xt) based on the target topology graph G1 and the multi-channel time series data, input the graph dataset into the graph neural network model, and obtain the fault propagation probability matrix;

[0009] Identifying a suspected fault area according to the fault propagation probability matrix, synchronously injecting pulse signals into nodes in the suspected fault area, and collecting node response signals;

[0010] The fault location of the distribution network is located according to the fault propagation probability matrix and the node response signal.

[0011] Preferably, the incremental update module includes:

[0012] The event perception layer is used to obtain topology change data of the distribution network; the topology change data includes switch state change data, distributed energy grid connection and disconnection switching data, load switching action data, and relay protection action data;

[0013] The event processing layer is used to parse the topology change data obtained by the event perception layer, obtain the node where the topology change event is located and the event type, and trigger a topology update when a topology change event occurs;

[0014] The topology update layer is used to traverse the topology database to identify the node set and edge set affected by the node where the topology change event is located, obtain the incremental node set ΔV and the incremental edge set ΔE, and update the initial topology graph based on the following method: G1=(V0∪ΔV, E0∪ΔE); V0∪ΔV represents the union of the initial node set and the incremental node set; E0∪ΔE represents the union of the initial edge set and the incremental edge set.

[0015] Preferably, traversing the topology database includes: taking the node where the topology change event is located as the starting node, traversing the adjacent nodes layer by layer from the starting node along the connection path based on a graph traversal algorithm to obtain the incremental node set ΔV and the incremental edge set ΔE;

[0016] The traversal is terminated when any of the following conditions are met:

[0017] The accumulated path resistance reaches the threshold resistance;

[0018] The number of hops between the current node and the starting node exceeds the set number of hops.

[0019] Preferably, updating the initial topology graph G0 to obtain the target topology graph G1 includes adding edges or deleting edges, and the newly added edges include: searching for the physical line corresponding to the newly added edge from the line parameter library, and extracting the resistance value and reactance value of the physical line; calculating the electrical impedance parameters according to the resistance value and reactance value, and assigning weights to the newly added edges in the target topology graph G1 according to the electrical impedance parameters.

[0020] Preferably, the deleting of edges includes: determining whether there is a reachable path between the power supply node and the target load node in the target topology graph G1; if a topology break is detected, marking the broken link area as an island subgraph.

[0021] Preferably, the construction graph dataset H=(G1, Xt) includes mapping the multi-channel time series data Xt to the nodes of the target topology graph G1 to obtain the node feature matrix X; establishing the adjacency matrix A of the target topology graph G1, and combining the adjacency matrix A of the target topology graph G1 with the node feature matrix X to obtain the construction graph dataset H.

[0022] Preferably, the graph neural network model includes:

[0023] A graph convolution layer is used to perform a graph convolution operation on the graph dataset H to obtain an intermediate feature vector reflecting the influence of the adjacent node status;

[0024] An attention mechanism layer, connected to the output of the graph convolution layer, for performing attention enhancement on the intermediate feature vector to obtain an attention-enhanced node feature vector;

[0025] The fully connected layer is used to map the attention-enhanced node feature vector into a fault propagation probability to obtain the fault propagation probability matrix.

[0026] Preferably, performing attention enhancement on the intermediate feature vector includes weighting the intermediate feature vector based on a connection weight, wherein the connection weight is determined based on the following method: determining a physical line corresponding to a connection edge between two nodes, and extracting a resistance value and a reactance value of the physical line;

[0027] Calculating electrical impedance parameters according to the resistance value and reactance value; and assigning weights to the connection edges of the two nodes according to the electrical impedance parameters to obtain the connection weights.

[0028] Preferably, the pulse signal is synchronously injected into k nodes with the highest probability of fault propagation in the suspected fault area, forming multi-node synchronous injection, k≥2; wherein the frequency of the pulse signal is set to 10kHz~50kHz.

[0029] In a second aspect, to solve the above technical problems, the present invention provides a distribution network fault location system, comprising:

[0030] The topology construction module is used to construct an initial topology graph G0 = (V0, E0) based on the power equipment of the distribution network and their connection relationships. V0 represents a node set; E0 represents an edge set. The power equipment is mapped as a node; the connection relationship between the power equipment is mapped as an edge.

[0031] A topology incremental update module is configured to, when a topology change event occurs, identify an incremental node set and an incremental edge set affected by the topology change event, and update the initial topology graph G0 according to the incremental node set and incremental edge set to obtain a target topology graph G1;

[0032] A graph data construction module is used to construct a graph data set H = (G1, Xt) by combining the target topology graph G1 and the node multi-channel time series data Xt when the fault occurs;

[0033] A graph neural network module is used to process the graph dataset H to obtain a fault propagation probability matrix for each node in the distribution network;

[0034] a signal injection module, configured to identify a suspected fault area according to the fault propagation probability matrix, synchronously inject pulse signals into nodes within the suspected fault area, and collect node response signals;

[0035] The fault location module is used to locate the distribution network fault location according to the fault propagation probability matrix and the node response signal.

[0036] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0037] The distribution network fault location method and system described in the present invention, through dynamic topology updating, multi-source data fusion and physical signal verification, can quickly and accurately locate the fault location when the distribution network topology structure dynamically changes, significantly reduce the deviation rate and misjudgment rate of distribution network fault location, and improve the adaptability to complex fault scenarios.

[0038] Among them, the incremental update module is driven by topology change events, and the topology graph is only locally updated for the nodes and edges affected by the topology change events, avoiding the reconstruction of the entire graph, improving the efficiency of topology processing, and ensuring that the topology graph of the input graph neural network model is always synchronized with the operating status of the distribution network, solving the misjudgment caused by the lag of topology information.

[0039] Construct a graph dataset containing multi-channel time series data such as voltage and current and topological connection relationships, use graph convolution and attention mechanism to model the fault propagation path, fully capture the graph structure dependency of electrical disturbances between nodes, and improve the accuracy of identifying the fault scope and propagation path.

[0040] By synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting the response signals of all controlled nodes, the suspected fault area is verified in combination with the laws of physics, significantly reducing the false alarm rate.

[0041] By synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting node responses, high-resolution inference of the fault point coordinates is performed based on the propagation probability distribution and the frequency domain response characteristics of the controlled nodes. This significantly improves the accuracy of fault location, breaking through the limitation of only being able to locate the node and achieving sub-node level accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0043] Figure 1 Flowchart of a method for locating a distribution network fault in a preferred embodiment of the present invention;

[0044] Figure 2 A flow chart of determining connection weights in a preferred embodiment of the present invention;

[0045] Figure 3 2 is a structural block diagram of a distribution network fault location system in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0047] The purpose of the embodiments of the present invention is to solve the problem of fault location deviation or misjudgment caused by not considering topological changes including switch switching, distributed energy grid connection and disconnection switching, distributed energy load switching or relay protection action when locating distribution network faults.

[0048] Example 1: Reference Figure 1 As shown, an embodiment of the present invention discloses a distribution network fault location method, comprising:

[0049] S10. Construct an initial topology graph G0 = (V0, E0) of the distribution network; V0 represents a node set; E0 represents an edge set; wherein power equipment is mapped as a node; and the connection relationship between power equipment is mapped as an edge;

[0050] S20, triggering an incremental update module when a topology change event occurs, obtaining an incremental node set and an incremental edge set affected by the topology change event, and updating the initial topology graph G0 according to the incremental node set and incremental edge set to obtain a target topology graph G1;

[0051] S30, obtaining multi-channel time series data Xt of the node when the fault occurs, constructing a graph dataset H=(G1, Xt) according to the target topology graph G1 and the multi-channel time series data, inputting the graph dataset into a graph neural network model, and obtaining a fault propagation probability matrix;

[0052] S40, identifying a suspected fault area according to the fault propagation probability matrix, synchronously injecting a pulse signal into nodes in the suspected fault area, and collecting node response signals;

[0053] S50: Locate a distribution network fault location according to the fault propagation probability matrix and the node response signal.

[0054] The distribution network fault location method described in the present invention, through dynamic topology updating, multi-source data fusion and physical signal verification, can quickly and accurately locate the fault position when the distribution network topology structure dynamically changes, significantly reduce the deviation rate and misjudgment rate of distribution network fault location, and improve the adaptability to complex fault scenarios.

[0055] In a specific application scenario, step S10: By connecting to a supervisory control and data acquisition system (also known as a SCADA (Supervisory Control and Data Acquisition) system, a geographic information system (also known as a GIS (Geographic Information System)) system, or a distribution automation terminal (such as a feeder terminal unit (FTU) or a distribution transformer terminal unit (TTU)), information about power equipment in the distribution network is collected to construct a node set V0 and an edge set E0. Node set V0 includes power equipment such as transformers, circuit breakers, disconnectors, and feeder switches as nodes in the graph. Edge sets are generated based on the actual connections between the devices. Each edge corresponds to a physical line segment, and its attributes include resistance, reactance, impedance, line length, and rated current. The initial topology graph is a structured input that reflects the electrical connection status of the distribution network at the initial moment. It is represented in the form of an adjacency matrix, an edge table, and a graph database to support subsequent incremental updates and graph neural network model input.

[0056] Step S20: When the SCADA system detects a topology change event (such as a switch status change, feeder switching, or equipment decommissioning), it automatically triggers the incremental update module. Starting at the node where the topology change event occurred, it traverses the affected adjacent nodes to extract the incremental node set ΔV and incremental edge set ΔE. G1 = (V0∪ΔV, E0∪ΔE) is then executed on the initial topology graph, where V0∪ΔV represents the union of the initial node set and the incremental node set; and E0∪ΔE represents the union of the initial edge set and the incremental edge set. The resulting target topology graph G1 represents the current real-time electrical topology of the distribution network. By driving the incremental update module with topology change events, the topology graph is only partially updated for the nodes and edges affected by the topology change event, avoiding full graph reconstruction and improving topology processing efficiency. This ensures that the topology graph input to the graph neural network model is always synchronized with the operating status of the distribution network, resolving misjudgments caused by lagging topology information.

[0057] Step S30: When a fault event is detected, multi-channel time-series electrical quantity data (referred to as "multi-channel time-series data Xt") is synchronously acquired for each node, including the three-phase voltage amplitude and phase angle, the three-phase current amplitude and phase angle, short-time Fourier transform characteristics or transient disturbance response, etc. This multi-channel time-series data Xt is bound to the target topology graph to form a graph dataset H. This graph dataset H is input into the graph neural network model, and a fault propagation probability matrix is ​​output. This matrix reflects the degree of impact of each node under the current topology and fault state, and serves as a basis for identifying suspected fault areas. By constructing a graph dataset containing the topological connection relationship between multi-channel time-series data such as voltage and current, and graph convolution and attention mechanisms to model the fault propagation path, the graph structure dependencies of electrical disturbances between nodes are fully captured, improving the accuracy of identifying the fault impact range and propagation path.

[0058] Step S40: Nodes with high values ​​in the fault propagation probability matrix are divided into suspected fault areas. Pulse signals are synchronously injected into all nodes in the suspected fault area according to the injection strategy. The response signals of all controlled nodes within the injection window are collected through the FTU. Time domain or frequency domain features are extracted from the response signals, such as amplitude change rate, spectrum offset, arrival event, etc. The data obtained by feature processing is used to invert the spatial position of the fault point.

[0059] Step S50: By synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting the response signals of all controlled nodes within the pulse signal injection window, a joint optimization objective function is established to infer the fault location by combining the propagation probability distribution and the frequency domain response characteristics, and finally the coordinates of the fault location are determined to support sub-node level fault location.

[0060] Specifically, in the above-mentioned scheme of the embodiment of the present invention, the incremental update module is the core key part of the distribution network fault location, which is used to quickly and accurately update the power grid topology map when the distribution network structure changes. The incremental update module includes an event perception layer, an event processing layer and a topology update layer. The three work together to realize the transition from "perceiving topology changes" to "incremental map updates".

[0061] Among them, the event perception layer is used to obtain topology change data of the distribution network; the topology change data includes switch state change data, distributed energy grid connection and disconnection switching data or load switching action data, and relay protection action data.

[0062] In specific application scenarios, the event perception layer establishes a data subscription channel with the SCADA system or smart terminals to receive and aggregate topology change data from the SCADA system. This includes switch status change data (such as the opening and closing of circuit breakers and load switches); distributed energy resource on-grid switching data (such as the on-grid and off-grid switching of photovoltaic inverters); load switching action data (such as the switching of load-side switches on transformers); and relay protection action data (such as overcurrent protection, tripping, and reclosing). By setting event triggering rules (such as status bit changes and telesignaling exceeding limits), the topology change event trigger stamp corresponding to the device number is collected. These events enable automatic identification and real-time perception of grid topology changes.

[0063] The event processing layer is used to parse the topology change data obtained by the event perception layer, obtain the node where the topology change event is located and the event type, and trigger a topology update when a topology change event occurs.

[0064] In specific application scenarios, the event processing layer performs structured processing on the topology change data obtained by the event perception layer, extracts the source node, device ID and action type of the topology change event, determines the type of event (such as disconnection / closure / on-grid connection, etc.) based on the rule engine, and determines whether the event affects the network topology structure based on the event impact rules. If it is a topology change-related event, it enters the topology update layer. Through event parsing, event classification, update judgment and trigger control, invalid events are avoided to cause redundant calculations, ensuring that updates are triggered only when necessary.

[0065] The topology update layer is used to traverse the topology database to identify the node set and edge set affected by the node where the topology change event is located, obtain the incremental node set ΔV and incremental edge set ΔE, and update the initial topology graph based on the following method: G1=(V0∪ΔV, E0∪ΔE); V0∪ΔV represents the union of the initial node set and the incremental node set; E0∪ΔE represents the union of the initial edge set and the incremental edge set.

[0066] In specific application scenarios, based on the node where the topology change event occurs, the adjacency table / topology database is called to identify its connection relationships. The newly connected or invalid nodes affected by the topology change event are obtained as the incremental node set, and the newly connected or disconnected lines connected to the incremental nodes are obtained as the incremental edge set. Based on the topology change information provided by the event processing layer, the topology update layer performs local incremental modifications to the initial topology graph, avoiding full graph reconstruction and improving operational efficiency and stability. It also supports the flexible evolution and continuous updating of the topology structure to meet the requirements of dynamic access to distributed power sources.

[0067] Specifically, traversing the topology database includes starting at the node where the topology change event occurs, and traversing the adjacent nodes layer by layer from the starting node along the connection path based on a graph traversal algorithm to obtain the incremental node set ΔV and incremental edge set ΔE. Graph traversal algorithms, such as breadth-first traversal (BFS) or depth-first traversal (DFS) graph algorithms, start from the starting node and expand adjacent nodes and edges layer by layer according to the topological graph connection relationship, and dynamically record the resistance and hop count of the current path. Each time a node is visited, the cumulative path resistance and hop count between it and the starting node are determined. The traversal process terminates when any of the following conditions are met, limiting the topology update to irrelevant areas:

[0068] Condition 1: The accumulated path resistance reaches the threshold resistance.

[0069] This means that when the cumulative path resistance from the starting node to the previous node is greater than or equal to the threshold resistance, expansion in that direction is terminated. Because higher resistance indicates greater distance and weaker coupling, the impact of topological changes on alignment can be ignored. The threshold resistance can be set empirically, for example, at 1.5 to 2 times the average impedance between the trunk and branch lines. Typical reference values ​​include 1.5 to 3.5 ohms for a 10kV regional network and 0.8 to 2.5 ohms for a 35kV industrial network.

[0070] Condition 2: The number of hops between the current node and the starting node exceeds the set number of hops.

[0071] This means that if the number of hops between the current node and the starting node exceeds the set number of hops, further expansion will cease. The set number of hops can be determined based on a topology approach. For example, a setting of 3 to 5 layers generally covers station-line-station areas. Alternatively, the set number of hops can be adjusted based on event type mapping, with different event types requiring different hops (e.g., 2 hops for a connection or disconnection, 4 hops for a trip).

[0072] After the traversal is completed, all visited nodes constitute the incremental node set ΔV, and the edges actually traversed constitute the incremental edge set ΔE.

[0073] When updating the topology, each edge represents a physical line. Its electrical characteristics (resistance, reactance, and impedance) directly affect electrical propagation and fault diagnosis. Ignoring these characteristics and treating them as simple connections can lead to serious modeling errors and diagnostic biases. For example, fault propagation paths can be distorted, where a high-resistance line with very weak propagation is mistakenly identified as the primary path, reinforcing the incorrect path. Furthermore, the inability to distinguish between near and far neighbors can be a problem. In distribution networks, topological proximity does not necessarily equate to electrical proximity. If line resistance and reactance are not considered, two edges in the diagram may have the same length, but electrically one may have a resistance of only 0.2 ohms and the other as high as 5 ohms. Failure to model this difference can lead to the "far neighbor" being mistaken for the "near neighbor," causing the suspected fault area to expand or shift.

[0074] To address this issue, in an embodiment of the present invention, updating the initial topology to obtain the target topology includes adding or deleting edges. The adding of edges includes: searching for the physical line corresponding to the added edge from the line parameter library and extracting the resistance and reactance values ​​of the physical line; calculating the electrical impedance parameters based on the resistance and reactance values, and assigning weights to the added edges in the target topology G1 based on the electrical impedance parameters. By calling the line parameter library to annotate the resistance and reactance of the added edges, the impedance is calculated and written into the target topology G1 as the weight of the added edge, making the target topology G1 a dynamic data carrier that integrates the two dimensions of structure and electricity, ensuring that the edge weights are consistent with the actual distribution network structure.

[0075] After the distribution network topology changes, deleting edges may cause some loads to lose connection with the power supply, resulting in topological breaks or unplanned island operation. If not identified in time, it will mislead fault diagnosis, power flow scheduling, and even cause safety accidents. In order to solve this problem, in the embodiment of the present invention, a connectivity analysis and island detection mechanism is performed immediately after the edge is updated: the edge deletion includes: determining whether there is a reachable path between the power supply node and the target load node in the target topological graph, and if a topological break is detected, the broken link area is identified as an island subgraph. Here, the island subgraph identification logic is: if there is a target load node that is not in the reachable set of any power supply node, it means that it has been disconnected from the power supply, and the subgraph area where the node is located is identified as an island subgraph.

[0076] When constructing the input for the graph neural network model, the updated target topology graph is combined with the multi-channel time series data of each node to form a graph dataset with dual structural and state characteristics. Specifically, the graph dataset H = (G1, Xt) is constructed, including mapping the multi-channel time series data Xt to the nodes of the target topology graph G1 to obtain the node feature matrix X; the adjacency matrix A of the target topology graph G1 is established, and the constructed graph dataset H is obtained by combining the adjacency matrix A of the target topology graph G1 with the node feature matrix X. The process of constructing the graph dataset H = (G1, Xt) deeply integrates structural information with operational state data, providing the subsequent graph neural network model with complete input with electrical and physical significance, temporal evolution capabilities, and edge weight propagation logic.

[0077] The graph neural network model includes a graph convolution layer, an attention mechanism layer and a fully connected layer. The graph convolution layer performs a graph convolution operation on the graph dataset H to obtain an intermediate feature vector reflecting the influence of the adjacent node status; the attention mechanism layer connects the output of the graph convolution layer and performs attention enhancement on the intermediate feature vector, including weighting the degree of influence of different adjacent nodes to obtain an attention-enhanced node feature vector and strengthen the critical path; the fully connected layer maps the attention-enhanced node feature vector to a fault propagation probability to obtain the fault propagation probability matrix.

[0078] In specific application scenarios, the graph convolution layer uses a multi-layer stacking structure (such as 2~3 layers of GCN) to cover multi-hop neighborhoods; the attention mechanism layer adopts a multi-head attention mechanism, supports edge attributes to participate in attention scoring, performs weighted aggregation on the intermediate features output by the graph convolution layer, learns the influence of different neighbor nodes on the current node, and improves the key path weight.

[0079] Further, performing attention enhancement on the intermediate feature vector includes weighting the intermediate feature vector based on the connection weight, referring to Figure 2 As shown, the connection weight is determined based on the following method: determining the physical line corresponding to the connection edge of two nodes, and extracting the resistance value and reactance value of the physical line; calculating the electrical impedance parameters according to the resistance value and reactance value; assigning weights to the connection edge of the two nodes according to the electrical impedance parameters to obtain the connection weight.

[0080] In specific application scenarios, electrical impedance parameters are mapped to connection weights. Larger electrical impedance values ​​indicate weaker electrical coupling between two nodes, lower propagation capabilities, and smaller connection weights. Main branches with smaller electrical impedance values ​​receive higher attention weights and are more likely to form clusters of nodes with high fault probabilities. In the graph attention mechanism, the physical properties of edges are used as key influencing factors in the attention calculation process. By extracting the line impedance characteristics represented by the edges and mapping them to connection strengths (weights), the attention mechanism is guided to perform weighted aggregation on adjacent nodes. This implements a physically constrained graph attention learning mechanism, addressing modeling errors and diagnostic biases. This results in more accurate and consistent fault location results in complex distribution networks (such as ring networks / distributed energy access). Furthermore, in scenarios with frequent structural changes, physical edge weights serve as a stable prior to enhance the model's anti-interference capabilities.

[0081] Furthermore, the pulse signal is synchronously injected into the k nodes with the highest probability of fault propagation within the suspected fault area, forming a multi-node synchronous injection, where k ≥ 2. Multi-point injection simultaneously covers multiple propagation paths, preventing single-point injection from missing remote branches or obstructing paths. Furthermore, if an injection point has a local impedance anomaly or poor grounding, other injection points can still receive an effective response, improving fault tolerance. Furthermore, multi-point injection generates interference and combined waveforms, which helps improve the model's recognition capabilities during inversion optimization.

[0082] Among them, the pulse signal is a high-frequency pulse signal. The high-frequency pulse signal has high electromagnetic coupling sensitivity and can penetrate more branch nodes. The signal is more obviously affected by electrical impedance, branch topology, and local defects, and is more easily reflected or absorbed by the fault point, which facilitates fault point identification.

[0083] Specifically, the frequency of the high-frequency pulse signal is set to 10kHz~50kHz, and in addition to having the characteristics of a high-frequency pulse signal, it will not interfere with the normal power supply of the low-frequency power grid.

[0084] Specifically, the fault location of the distribution network is located according to the fault propagation probability matrix and the node response signal, including establishing a joint optimization objective function to infer the fault location by combining the fault propagation probability distribution and the frequency domain response characteristics. The specific implementation process is as follows:

[0085] Process the node response signal R(t) of the suspected fault area to obtain the frequency domain characteristics of the node response signal; assume that the fault occurs at position x and define the objective function, and solve the objective function to minimize the fault position x; the objective function is:

[0086] ;

[0087] P(x) represents the probability of the simulated fault signal propagating to each node in the distribution network topology, assuming that the fault occurs at location x; represents the fault propagation probability output by the graph neural network model; represents the balance coefficient; S(x) represents the frequency domain characteristics of the node response caused by the simulated pulse signal propagating from the fault location x to other nodes in the distribution network topology; R(f) represents the actual measured frequency domain characteristics of the node response.

[0088] By fitting theoretical propagation behavior with actual observations, the fault location problem is transformed into an optimal matching problem. The position that minimizes the overall propagation error is determined as the fault location x. Therefore, by synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting node responses, a joint optimization objective function is established by combining the propagation probability distribution and frequency domain response characteristics. This enables high-resolution inference of the fault point coordinates, significantly improving the granularity of fault location, breaking through the limitation of locating only at the node level and achieving sub-node-level accuracy.

[0089] Example 2: This embodiment of the present invention is based on the same inventive concept as Example 1. Figure 3 As shown, a distribution network fault location system is provided, comprising:

[0090] The topology construction module is used to construct an initial topology graph G0 = (V0, E0) based on the power equipment of the distribution network and their connection relationships. V0 represents a node set; E0 represents an edge set. The power equipment is mapped as a node; the connection relationship between the power equipment is mapped as an edge.

[0091] A topology incremental update module is configured to, when a topology change event occurs, identify an incremental node set and an incremental edge set affected by the topology change event, and update the initial topology graph G0 according to the incremental node set and incremental edge set to obtain a target topology graph G1;

[0092] A graph data construction module is used to construct a graph data set H = (G1, Xt) by combining the target topology graph G1 and the node multi-channel time series data Xt when the fault occurs;

[0093] A graph neural network module is used to process the graph dataset H to obtain a fault propagation probability matrix for each node in the distribution network;

[0094] a signal injection module, configured to identify a suspected fault area according to the fault propagation probability matrix, synchronously inject pulse signals into nodes within the suspected fault area, and collect node response signals;

[0095] The fault location module is used to locate the distribution network fault location according to the fault propagation probability matrix and the node response signal.

[0096] The embodiment of the present invention and the first embodiment are based on the same inventive concept and have the same technical effects, which will not be described in detail here.

[0097] In summary, the distribution network fault location method and system described in the present invention, through dynamic topology updating, multi-source data fusion and physical signal verification, can quickly and accurately locate the fault location when the distribution network topology structure dynamically changes, significantly reduce the deviation rate and misjudgment rate of distribution network fault location, and improve the adaptability to complex fault scenarios.

[0098] Among them, the incremental update module is driven by topology change events, and the topology graph is only locally updated for the nodes and edges affected by the topology change events, avoiding the reconstruction of the entire graph, improving the efficiency of topology processing, and ensuring that the topology graph of the input graph neural network model is always synchronized with the operating status of the distribution network, solving the misjudgment caused by the lag of topology information.

[0099] Construct a graph dataset containing multi-channel time series data such as voltage and current and topological connection relationships, use graph convolution and attention mechanism to model the fault propagation path, fully capture the graph structure dependency of electrical disturbances between nodes, and improve the accuracy of identifying the fault scope and propagation path.

[0100] By synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting the response signals of all controlled nodes, the suspected fault area is verified in combination with the laws of physics, significantly reducing the false alarm rate.

[0101] By synchronously injecting pulse signals into multiple nodes in the suspected fault area and collecting node responses, high-resolution inference of the fault point coordinates is performed based on the propagation probability distribution and the frequency domain response characteristics of the controlled nodes. This significantly improves the accuracy of fault location, breaking through the limitation of only being able to locate the node and achieving sub-node level accuracy.

[0102] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0104] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0106] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A distribution network fault location method, characterized by: include, Construct the initial topology graph of the distribution network G0 = (V0, E0); V0 represents the node set; E0 represents the edge set; among them, power equipment is mapped as a node; the connection relationship between power equipment is mapped as an edge; When a topology change event occurs, the incremental update module is triggered to obtain an incremental node set and an incremental edge set affected by the topology change event, and the initial topology graph G0 is updated according to the incremental node set and the incremental edge set to obtain a target topology graph G1; Obtain multi-channel time series data Xt of the node when the fault occurs, construct a graph dataset H=(G1, Xt) based on the target topology graph G1 and the multi-channel time series data, input the graph dataset into the graph neural network model, and obtain the fault propagation probability matrix; Identifying a suspected fault area according to the fault propagation probability matrix, synchronously injecting pulse signals into nodes in the suspected fault area, and collecting node response signals; Locating the distribution network fault location according to the fault propagation probability matrix and the node response signal; The incremental update module includes: The event perception layer is used to obtain topology change data of the distribution network; the topology change data includes switch state change data, distributed energy grid connection and disconnection switching data, load switching action data, and relay protection action data; The event processing layer is used to parse the topology change data obtained by the event perception layer, obtain the node where the topology change event is located and the event type, and trigger a topology update when a topology change event occurs; The topology update layer is used to traverse the topology database to identify the node set and edge set affected by the node where the topology change event occurs, obtain the incremental node set ΔV and incremental edge set ΔE, and update the initial topology graph based on the following method: G1=(V0∪ΔV, E0∪ΔE).

2. The distribution network fault location method according to claim 1, characterized in that: Traversing the topology database includes taking the node where the topology change event occurs as a starting node, and traversing adjacent nodes layer by layer from the starting node along the connection path based on a graph traversal algorithm to obtain the incremental node set ΔV and the incremental edge set ΔE; The traversal is terminated when any of the following conditions are met: The accumulated path resistance reaches the threshold resistance; The number of hops between the current node and the starting node exceeds the set number of hops.

3. The distribution network fault location method according to claim 1, characterized in that: Updating the initial topology graph G0 to obtain the target topology graph G1 includes adding edges or deleting edges, and the adding edges includes: searching for a physical line corresponding to the added edge from a line parameter library, and extracting the resistance value and reactance value of the physical line; calculating electrical impedance parameters based on the resistance value and reactance value, and assigning weights to the added edges in the target topology graph G1 based on the electrical impedance parameters.

4. The distribution network fault location method according to claim 3, characterized in that: The deleting of the edge includes: determining whether there is a reachable path between the power supply node and the target load node in the target topology graph G1; if a topology break is detected, marking the broken link area as an island subgraph.

5. The distribution network fault location method according to claim 1, characterized in that: Construct graph dataset H=(G1,Xt), including, Mapping the multi-channel time series data Xt to the nodes of the target topology graph G1 to obtain the node feature matrix X; An adjacency matrix A of the target topology graph G1 is established, and the construction graph dataset H is obtained by combining the adjacency matrix A of the target topology graph G1 with the node feature matrix X.

6. The distribution network fault location method according to claim 1, characterized in that: The graph neural network model includes: A graph convolution layer is used to perform a graph convolution operation on the graph dataset H to obtain an intermediate feature vector reflecting the influence of the adjacent node status; An attention mechanism layer, connected to the output of the graph convolution layer, for performing attention enhancement on the intermediate feature vector to obtain an attention-enhanced node feature vector; The fully connected layer is used to map the attention-enhanced node feature vector into a fault propagation probability to obtain the fault propagation probability matrix.

7. The distribution network fault location method according to claim 6, characterized in that: Performing attention enhancement on the intermediate feature vector includes weighting the intermediate feature vector based on a connection weight, wherein the connection weight is determined based on: Determine the physical line corresponding to the connecting edge of the two nodes, and extract the resistance and reactance values ​​of the physical line; Calculating electrical impedance parameters based on the resistance value and reactance value; Weights are assigned to the connection edges of the two nodes according to the electrical impedance parameters to obtain the connection weights.

8. The distribution network fault location method according to claim 1, characterized in that: The pulse signal is synchronously injected into k nodes with the highest probability of fault propagation in the suspected fault area, forming multi-node synchronous injection, k≥2; wherein the frequency of the pulse signal is set to 10kHz~50kHz.

9. Distribution network fault location system, characterized by: include, The topology construction module is used to construct an initial topology graph G0 = (V0, E0) based on the power equipment of the distribution network and their connection relationships. V0 represents a node set; E0 represents an edge set. The power equipment is mapped as a node; the connection relationship between the power equipment is mapped as an edge. A topology incremental update module is configured to, when a topology change event occurs, identify an incremental node set and an incremental edge set affected by the topology change event, and update the initial topology graph G0 according to the incremental node set and incremental edge set to obtain a target topology graph G1; A graph data construction module is used to construct a graph data set H = (G1, Xt) by combining the target topology graph G1 and the node multi-channel time series data Xt when the fault occurs; A graph neural network module is used to process the graph dataset H to obtain a fault propagation probability matrix for each node in the distribution network; a signal injection module, configured to identify a suspected fault area according to the fault propagation probability matrix, synchronously inject pulse signals into nodes within the suspected fault area, and collect node response signals; A fault location module is used to locate the fault position of the distribution network according to the fault propagation probability matrix and the node response signal; The incremental update module includes: The event perception layer is used to obtain topology change data of the distribution network; the topology change data includes switch state change data, distributed energy grid connection and disconnection switching data, load switching action data, and relay protection action data; The event processing layer is used to parse the topology change data obtained by the event perception layer, obtain the node where the topology change event is located and the event type, and trigger a topology update when a topology change event occurs; The topology update layer is used to traverse the topology database to identify the node set and edge set affected by the node where the topology change event occurs, obtain the incremental node set ΔV and incremental edge set ΔE, and update the initial topology graph based on the following method: G1=(V0∪ΔV, E0∪ΔE).

Images