Power distribution network fault positioning method, apparatus and device, and storage medium

Abstract

The invention discloses a power distribution network fault positioning method, device and equipment and a storage medium, which are applied to the field of power distribution network fault positioning, and the method comprises the steps: constructing a symmetric node adjacency relation data set and an electrical parameter data set; inputting the two data sets into a first fault prediction model to carry out preliminary fault prediction, and analyzing an association relationship among fault nodes in a prediction result; performing dimension reduction processing on the original electrical parameters to obtain a simplified feature set, and inputting the simplified feature set and the node association information into a second graph neural network model to obtain deep association information; and finally, integrating the preliminary prediction result, the dimensionality-reduced feature set and the deep correlation information by using a graph convolutional network to obtain a fault prediction result, and generating a detailed fault positioning log. According to the power distribution network fault positioning method, device and equipment and the storage medium provided by the invention, the positioning accuracy of the power distribution network fault is improved, and a technical guarantee is provided for stable operation of the power distribution network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network fault positioning, in particular to a power distribution network fault positioning method, device, equipment and storage medium. BACKGROUND

[0002] With the continuous development of the power distribution network as an important infrastructure in the power system, reliable, fast and accurate fault positioning of the power distribution network can provide strong decision basis for the operation management department, greatly reduce the operation cost of the power distribution network and improve the reliability and safety of the power system.

[0003] The existing fault positioning technology uses a machine learning model to detect and identify faults, for example, an ANN (Artificial Neural Network) model. However, this method relies heavily on the number of sample data, but the number of labeled data in the sample data is limited, which leads to a decrease in the accuracy of the model in identifying power distribution network faults under low label rate conditions, and further leads to difficulty in accurately positioning the power distribution network faults, which further causes serious equipment damage or power outage accidents and further affects the work progress of the entire power distribution network. SUMMARY

[0004] The present application provides a power distribution network fault positioning method, device and storage medium to solve the technical problem of power distribution network fault positioning and to achieve the effect of improving the accuracy of power distribution network fault positioning.

[0005] To solve the above technical problems, the present application provides a power distribution network fault positioning method, device and storage medium. The method comprises: Obtaining node topology information of a target power distribution network to obtain an information-symmetrical power grid node adjacency relationship dataset; Obtaining operation information of a selected target node to obtain a first power grid electrical parameter dataset; Respectively inputting the first power grid electrical parameter dataset and the information-symmetrical power grid node adjacency relationship dataset into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain a first fault prediction result of the target power distribution network; Analyzing the first fault prediction result to obtain first correlation information between each fault node corresponding to the first fault prediction result; Performing dimensionality reduction processing on the first power grid electrical parameter dataset to obtain a second power grid electrical parameter dataset; Respectively inputting the correlation information and the second power grid electrical parameter dataset into a second fault prediction model constructed by a graph neural network algorithm for processing to obtain second correlation information; inputting the first fault prediction result, the second power grid electrical parameter dataset and the second correlation information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target power distribution network; analyzing the second fault prediction result to generate a fault positioning log of the target power distribution network.

[0006] As one of the preferred solutions, the node topology information of the target power distribution network is obtained to obtain a power grid node adjacency relationship dataset with symmetric information, which includes: analyzing the node topology information of the target power distribution network to calculate the shortest path between nodes to obtain shortest path information; obtaining the power grid node adjacency relationship dataset based on the node topology information and the shortest path information of the target power distribution network; transforming the power grid node adjacency relationship dataset into a power grid node adjacency relationship dataset with symmetric information by using a maximum value transformation function.

[0007] As one of the preferred solutions, the first power grid electrical parameter dataset and the power grid node adjacency relationship dataset with symmetric information are respectively inputted into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain a first fault prediction result of the target power distribution network, which includes: performing an aggregation operation on the first power grid electrical parameter dataset and the power grid node adjacency relationship dataset with symmetric information to obtain network node embedding of the first fault prediction model constructed by the graph neural network algorithm; processing the network node embedding to obtain hidden features of nodes; converting the aggregated hidden features by using two fully connected layers to obtain the first fault prediction result of the target power distribution network; inputting into a fault probability preliminary prediction model adopting two fully connected layers and a cross-entropy loss function for layer-by-layer aggregation and probability conversion processing to obtain the first fault prediction result of the target power distribution network.

[0008] As one of the preferred solutions, the first fault prediction result at least includes each first fault node and a fault probability corresponding to each first fault node; and the first fault prediction result is analyzed to obtain first correlation information between each fault node corresponding to the first fault prediction result, which includes: based on the fault probability comparison result, screening each first fault node to obtain a second fault node; determining a fault event corresponding to each second fault node; extracting event features of each fault event; The first correlation information is obtained by a vector cosine distance of each event feature.

[0009] As one of the preferred solutions, the respective input of the correlation information and the second power grid electrical parameter data set into the second fault prediction model constructed by the graph neural network algorithm for processing to obtain the second correlation information includes: The second power grid electrical parameter data set is input into the second fault prediction model constructed by the graph neural network algorithm to obtain the graph learning weight of the second power grid electrical parameter data set. The graph learning weight and the correlation information are multiplied to obtain the second correlation information.

[0010] As one of the preferred solutions, the input of the first fault prediction result, the second power grid electrical parameter data set and the second correlation information into the third fault prediction model constructed by the graph convolution network algorithm to obtain the second fault prediction result of the target distribution network includes: The first fault prediction result, the second power grid electrical parameter data set and the second correlation information are input into the graph convolution layer of the third fault prediction model for layer-by-layer convolution to obtain a convolution result. The convolution result is input into a fully connected layer to obtain a fully connected output result. The fully connected output result is converted to obtain a second fault location prediction probability result.

[0011] As one of the preferred solutions, the analysis of the second fault prediction result to generate the fault positioning log of the target distribution network includes: The second fault prediction result is analyzed to obtain the probability of each node of the second fault prediction result to occur the fault event; The probability information is screened to determine the fault positioning. Based on the fault positioning, the fault positioning log of the target distribution network is generated.

[0012] Another aspect of the present application provides a distribution network fault positioning device, comprising: A topology information acquisition module is configured to acquire node topology information of a target distribution network to obtain an information-symmetrical power grid node adjacency relationship data set. An operation information acquisition module is configured to acquire operation information of a selected target node to obtain a first power grid electrical parameter data set. A first fault prediction model module is configured to respectively input the first power grid electrical parameter data set and the information-symmetrical power grid node adjacency relationship data set into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain a first fault prediction result of the target distribution network. an analysis module configured to analyze the first fault prediction result to obtain first association information between respective fault nodes corresponding to the first fault prediction result; a dimension reduction module configured to perform dimension reduction processing on the first power grid electrical parameter dataset to obtain a second power grid electrical parameter dataset; a second fault prediction model module configured to input the association information and the second power grid electrical parameter dataset into a second fault prediction model constructed by a graph neural network algorithm for processing to obtain second association information; a third fault prediction model module configured to input the first fault prediction result, the second power grid electrical parameter dataset and the second association information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target distribution network; a resolution module configured to resolve the second fault prediction result to generate a fault location log of the target distribution network.

[0013] In another aspect, the present application provides a distribution network fault location device, comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the distribution network fault location method as described above when executing the computer program.

[0014] In still another aspect, the present application provides a computer readable storage medium storing a computer program, wherein the computer readable storage medium implements the distribution network fault location method as described above when a device in which the computer readable storage medium is located executes the computer program.

[0015] The present application has at least one of the following advantages: The present application obtains node topology information of a target distribution network to obtain a power grid node adjacency relationship dataset with symmetric information, obtains operation information of a target node to obtain a first power grid electrical parameter dataset, then inputs the first power grid electrical parameter dataset and the power grid node adjacency relationship dataset with symmetric information into a fault probability preliminary prediction model constructed by a graph neural network model for processing to obtain a first fault location prediction probability result, and then optimizes the first fault location prediction probability result, inputs the first fault location prediction probability result, a second power grid electrical parameter dataset obtained after dimension reduction and second fault association data obtained after processing into a multi-layer topology-aware power grid fault location network composed of a graph convolution network to obtain a second fault location prediction probability result. According to the second fault location prediction probability result, the distribution network fault can be accurately located, the topology structure and electrical parameters are fused by the multi-layer graph neural network, the accuracy of the distribution network fault location is improved, and reliable technical support is provided for stable operation of the distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flow diagram of a power distribution network fault location method in one of the embodiments of the present application; Figure 2 is a schematic diagram of an observation node in one of the embodiments of the present application; Figure 3 is a structural diagram of a power distribution network fault location device in one of the embodiments of the present application; Figure 4 is a structural block diagram of a power distribution network fault location device in one of the embodiments of the present application; Figure 5 is a graphical structural diagram of an IEEE123 node test system in one of the embodiments of the present application; Reference signs: Among them, 301, processor; 302, memory; 201, obtain topological information module; 202, obtain running information module; 203, first fault prediction model module; 204, analysis module; 205, dimension reduction module; 206, second fault prediction model module; 207, third fault prediction model module; 208, analysis module; 1-114, the first type of node of power distribution network; 135, 150, 152, 160, 197, 250, 300, 350 and 450, the second type of node; 9r, 25r, 150r and 160r, voltage regulator; 61s, transformer; 451, close. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. The purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0018] In the description of the present application, the terms "first", "second", "third" and the like are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" and the like can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise stated, the meaning of "multiple" is two or more.

[0019] In the description of the present application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the art. The terms used in the specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. Those skilled in the art can understand the specific meanings of the above terms in the present application according to the specific circumstances.

[0020] It should be noted that the power distribution network is a key infrastructure of the power system, and its reliable operation is directly related to the energy supply of the social economy. With the popularity of distributed energy, the dynamics and complexity of the power distribution network have increased significantly, which puts higher requirements on fault location technology. Achieving fast and accurate fault diagnosis not only improves the safety and economy of power grid operation, but also provides effective decision support for management departments.

[0021] The method based on machine learning is subject to the problem of high cost of labeled data acquisition and sample scarcity, resulting in a significant decline in model performance at a low label rate.

[0022] An embodiment of the present application provides a power distribution network fault location method, device and storage medium, specifically, please refer to Figure 1 , Figure 1 The figure shows the flow chart of the power distribution network fault location method in one embodiment of the present application, the method comprises: S1, obtaining the node topology information of the target power distribution network, obtaining the information symmetric power grid node adjacency relationship dataset; S2, obtaining the operation information of the selected target node, obtaining the first power grid electrical parameter dataset; S3, respectively inputting the first power grid electrical parameter dataset and the information symmetric power grid node adjacency relationship dataset into the first fault prediction model constructed by the graph neural network algorithm for processing, obtaining the first fault prediction result of the target power distribution network; S4, analyzing the first fault prediction result to obtain the first correlation information between each fault node corresponding to the first fault prediction result; S5, performing dimensionality reduction processing on the first power grid electrical parameter dataset to obtain the second power grid electrical parameter dataset; S6, respectively inputting the correlation information and the second power grid electrical parameter dataset into the second fault prediction model constructed by the graph neural network algorithm for processing, obtaining the second correlation information; S7, inputting the first fault prediction result, the second power grid electrical parameter dataset and the second correlation information into the third fault prediction model constructed by the graph convolution network algorithm, obtaining the second fault prediction result of the target power distribution network; S8, analyze the second fault prediction result, and generate a fault location log of the target power distribution network.

[0023] In step S1, the node topology information of the target power distribution network is analyzed, the shortest path between nodes is calculated, and the shortest path information is obtained; based on the node topology information and the shortest path information of the target power distribution network, the power grid node adjacency relationship dataset is obtained; the power grid node adjacency relationship dataset is transformed into an information-symmetrical power grid node adjacency relationship dataset by using a maximum transformation function. Preferably, the power grid node adjacency relationship dataset can be represented as a topology-enhanced adjacency matrix, and the information-symmetrical power grid node adjacency relationship dataset can be represented as a symmetric topology-enhanced adjacency matrix. The topology-enhanced adjacency matrix is a matrix defined based on the shortest distance and the elements on the diagonal are 0, the element in the i-th row and the j-th column of the matrix is , , represents the average distance between the i-th node and its nearest nodes, is the shortest path between node i and node j, represents a set consisting of the nearest nodes to the i-th node; The topology-enhanced adjacency matrix is transformed to obtain a symmetric topology-enhanced adjacency matrix, and max() represents taking the maximum value. By symmetric processing, it is ensured that the power system is a physical topology of undirected graph in nature, which not only conforms to the actual connection relationship of the power network, but also ensures the bidirectional propagation of information between adjacent nodes, thereby avoiding prediction deviation caused by directional assumption.

[0024] In addition to the formula embodiments adopted in the foregoing embodiments, another embodiment of the present application also provides an alternative implementation path based on a machine learning model. Through a deep neural network architecture, data of a large-scale dataset is absorbed in the training stage, and multi-dimensional feature association and hidden rules are mastered. In actual deployment, only the input data needs to be imported into the trained model, and the built-in intelligent processing mechanism of the model can automatically generate high-precision output results meeting the requirements.

[0025] As one of the embodiments, the training method is to first construct a structured sample data set. The data set contains the basic input data of the node topology information of the target power distribution network, and also labels the matching target result label topology enhanced adjacency matrix matrix output, to establish the mapping relationship between the input and the output. Then, based on the data set, the initial AI model is trained by using the machine learning algorithm. During the training process, methods such as transfer learning, hyperparameter optimization or incremental training are used to optimize the model and enhance its generalization ability to unknown data. Finally, the trained optimized model matrix output model. In the deployment stage, the user only needs to input the node topology information of the target power distribution network to be analyzed to the model, and the model can automatically calculate and output the corresponding predicted result topology enhanced adjacency matrix through its learned internal law.

[0026] In step S2, the operation information of the selected target node is obtained to obtain a first power grid electrical parameter data set; preferably, the first power grid electrical parameter data set can be a power grid graph feature matrix. Preferably, the selected target node is an observation node, and details are described below. Figure 2 , Figure 2 The observation node shown in one of the embodiments of the application is shown in the schematic diagram, wherein 1-114 represent the first type of nodes of the power distribution network, 135, 150, 152, 160, 197, 250, 300, 350 and 450 represent the second type of nodes (the type is mainly determined by the installation position, model number and the like of the node, which is not specifically limited in the embodiment of the application), 9r, 25r, 150r and 160r represent voltage regulators, 61s represents a transformer, and 451 represents a switch. According to the structural characteristics of the power distribution network, the bus is set as a node, the connection relationship between the buses is regarded as an edge to construct a graph structure, and the three-phase voltage and phase of the bus are regarded as node characteristics, thereby realizing the conversion of the power distribution network data into an undirected graph G.

[0027] First, for an undirected graph G composed of N nodes, an observation node needs to be selected, the voltage and phase of the observation node are measurable, and the remaining nodes are non-observation nodes, the voltage and phase of the non-observation nodes are not measurable (both are 0). The observation nodes are not concentrated in one place, but are dispersed in various places of the power distribution network. A pair of nodes connected by a switch or a voltage regulator at the same position is regarded as a fault position where a fault may occur, and each of the remaining nodes is regarded as a fault position where a fault may occur, independently.

[0028] Suppose there are N nodes, then at any sampling point, the power grid graph feature matrix is constructed as: wherein, is the number of observed buses, is the data matrix of N nodes (including observation nodes and non-observation nodes, i.e. all nodes), T represents matrix transposition. The node data is , which represents the three-phase voltage and the phase of the node, wherein , respectively, represent the -phase, -phase, -phase voltage of the node. , respectively, represent the -phase, -phase, -phase phase angle of the node. For the i-th data sample, the feature matrix of the i-th data sample has the following form: wherein only the observation nodes have specific data, and the data of the non-observation nodes is 0. Through the aggregation and propagation of the electrical parameters of the key nodes, the state of the nodes not directly monitored is indirectly inferred, and the problem of insufficient observability caused by sparse sensor deployment is alleviated.

[0029] In addition to the formula embodiments adopted in the foregoing embodiments, another embodiment of the present application also provides an alternative implementation path based on a machine learning model. Through a deep neural network architecture, data of a large-scale data set is absorbed in the training stage, and multi-dimensional feature correlations and hidden rules are mastered. In actual deployment, only the input data needs to be imported into the trained model, and the high-precision output results meeting the requirements can be automatically generated through the built-in intelligent processing mechanism of the model.

[0030] As one of the embodiments, the training method is to first construct a structured sample data set. The data set contains basic input data observation node data , which represents the three-phase voltage and the phase of the node, wherein , respectively, represent the -phase, -phase, -phase voltage of the node. , respectively, represent the -phase, -phase, -phase phase angle of the node, and the target result label feature matrix output matching the same is also labeled to establish the mapping relationship between the input and the output. Subsequently, the initial AI model is trained based on the data set by using a machine learning algorithm. In the training process, the model is optimized by combining methods such as transfer learning, hyperparameter optimization or incremental training, so as to enhance its generalization ability to unknown data. Finally, the optimized model matrix output model is trained. In the deployment stage, the user only needs to input the observation node data to be analyzed into the model, and the model can automatically calculate and output the corresponding prediction result feature matrix through the internal rules learned by the model.​

[0031] In step S3, an aggregation operation is performed on the first power grid electrical parameter data set and the information-symmetrical power grid node adjacency relationship data set to obtain network node embedding of a first fault prediction model constructed by a graph neural network algorithm; the network node embedding is processed to obtain hidden features of nodes; and the aggregated hidden features are converted by using two full connection layers to obtain a first fault prediction result of the target distribution network. The first fault prediction result of the target distribution network is obtained by inputting the fault probability preliminary prediction model using two full connection layers and a cross-entropy loss function to layer-by-layer aggregation and probability conversion processing. As one of the embodiments, the node embedding of the first layer network of the graph neural network model is generated by aggregation based on a power grid graph feature matrix and a symmetric topology-enhanced adjacency matrix: The kth layer network aggregation of the graph neural network model includes: (1) Node sampling: obtaining node embedding of the current kth layer network (2) Node information aggregation: concatenating the current hidden features of the nodes with the aggregated neighbor features: (3) Node information update: inputting the concatenated features into a non-linear transformation such as a ReLU activation function to generate new hidden features of the nodes : (4) Through layer-by-layer aggregation of layers, the unobserved nodes can capture the features of the observed nodes within a multi-hop range, and then the final hidden features after K-layer aggregation are converted into prediction probabilities by using two full connections: In the process of training the graph neural network model, a cross-entropy loss function is used.

[0032] Further, in step S2, the graph neural network model converts the final hidden features after K-layer aggregation into prediction probabilities by using two full connections; in the process of training the graph neural network model, a cross-entropy loss function is used. Through topology-based information propagation, the voltage amplitude and phase features of the sparse observed nodes are converted into fault probability distribution of each node, so that the unobserved nodes can also obtain effective representation through the information of the adjacent observed nodes.

[0033] In step S4, the first fault prediction result at least includes each first fault node and a fault probability corresponding to each first fault node, each first fault node is screened based on the fault probability comparison result to obtain a second fault node; a fault event corresponding to each second fault node is determined; event features of each fault event are extracted; and the first association information is obtained by using vector cosine distance of each event feature. In one of the embodiments, the first association information is represented as a first feature association matrix.

[0034] Based on the first fault prediction result, the corresponding predicted fault position label is obtained according to the maximum value of the prediction probability, that is, the most likely fault node; With the actual power distribution network topology as a reference, the values corresponding to the nodes not directly connected to the most likely fault node are set to 0 to obtain new node embedding. The relevance between nodes is calculated according to the new node embedding, and the first feature correlation matrix is calculated using the cosine distance.

[0035] In step S6, the second power grid electrical parameter data set is input into a second fault prediction model constructed by a graph neural network algorithm to obtain graph learning weights of the second power grid electrical parameter data set; and the graph learning weights are multiplied by the correlation information to obtain second correlation information. Preferably, the second correlation information can be represented as a second feature correlation matrix. As one of the embodiments, the graph learning layer first generates the graph learning weights of the second power grid electrical parameter data set: Then, the first feature correlation matrix is multiplied by the graph learning weights to obtain the second feature correlation matrix.

[0036] As one of the embodiments, when the number of nodes increases, the calculation complexity of the graph structure increases. In order to reduce the calculation cost, a training matrix is used to reduce the dimensionality of the first power grid electrical parameter data set to obtain the first power grid electrical parameter data set, and the output of the graph learning layer is updated.

[0037] Compared with the matrix without graph learning, the second feature correlation matrix output by the graph learning layer can reduce noise interference, integrate weak supervision information, dynamically optimize the graph structure, and make the graph structure more suitable for label prediction requirements through loss function optimization.

[0038] In step S7, the first fault prediction result, the second power grid electrical parameter data set and the second correlation information are input into a graph convolution layer of a third fault prediction model for layer-by-layer convolution to obtain a convolution result; the convolution result is input into a fully connected layer to obtain a fully connected output result; and the fully connected output result is converted to obtain a second fault position prediction probability result.

[0039] Preferably, the graph convolution network includes L graph convolution layers and one fully connected layer, multiple graph convolution layers perform layer-by-layer convolution, and the fully connected layer converts the output of the last graph convolution layer into a prediction probability. In the training process, cross entropy is used as the loss function of the graph convolution network.

[0040] In step S8, the second fault prediction result is analyzed to generate a fault location log of the target power distribution network, including: analyzing the second fault prediction result to obtain a probability of each node of the second fault prediction result occurring the fault event; screening the probability information to determine fault location; and generating a fault location log of the target power distribution network based on the fault location. Preferably, the second fault prediction result can be a fault location prediction probability matrix, and fault location is determined by rows and columns of the final fault location prediction probability matrix. The matrix is an N*n matrix (N is the number of samples, and n is the number of power grid nodes), a row represents a sample (a fault event), and an output fault probability distribution of the event at all nodes is output. Each column corresponds to a node, and indicates a predicted probability that the node is a fault location.

[0041] Another embodiment of the application provides a power distribution network fault location device, please see Figure 3 , Figure 3 Fig. 1 shows a structural schematic diagram of a power distribution network fault location device according to an embodiment of the application, which includes: The topology information acquisition module 201 is configured to acquire node topology information of a target power distribution network to obtain an information-symmetrical power grid node adjacency relationship dataset. The operation information acquisition module 202 is configured to acquire operation information of a selected target node to obtain a first power grid electrical parameter dataset. The first fault prediction model module 203 is configured to input the first power grid electrical parameter dataset and the information-symmetrical power grid node adjacency relationship dataset into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain a first fault prediction result of the target power distribution network. The analysis module 204 is configured to analyze the first fault prediction result to obtain first correlation information between each fault node corresponding to the first fault prediction result. The dimension reduction module 205 is configured to perform dimension reduction processing on the first power grid electrical parameter dataset to obtain a second power grid electrical parameter dataset. The second fault prediction model module 206 is configured to input the correlation information and the second power grid electrical parameter dataset into a second fault prediction model constructed by a graph neural network algorithm for processing to obtain second correlation information. The third fault prediction model module 207 is configured to input the first fault prediction result, the second power grid electrical parameter dataset, and the second correlation information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target power distribution network. The analysis module 208 is configured to analyze the second fault prediction result to generate a fault location log of the target power distribution network.

[0042] Referring to Figure 4 which is a structural diagram of the power distribution network fault locating device provided by the embodiment of the present application, the power distribution network fault locating device 300 provided by the embodiment of the present application comprises a processor 301, a memory 302, and a computer program stored in the memory 302 and configured to be executed by the processor 301, wherein the processor 301 implements the steps in the power distribution network fault locating method embodiments as described above, such as steps S1-S8 in the embodiment described in the method embodiment, when executing the computer program; or the processor 301 implements the functions of the modules in each of the device embodiments, such as the topology information acquisition module 201, when executing the computer program. Figure 1

[0043] For example, the computer program can be divided into one or more modules, which are stored in the memory 302 and executed by the processor 301 to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the power distribution network fault locating device 300. For example, the computer program can be divided into the topology information acquisition module 201, the operation information acquisition module 202, the first fault prediction model module 203, the analysis module 204, the dimension reduction module 205, the second fault prediction model module 206, the third fault prediction model module 207, and the analysis module 208, and the specific functions of each module are as follows: The topology information acquisition module 201 is configured to acquire the node topology information of the target power distribution network to obtain the information-symmetrical power grid node adjacency relationship dataset; The operation information acquisition module 202 is configured to acquire the operation information of the selected target node to obtain the first power grid electrical parameter dataset; The first fault prediction model module 203 is configured to input the first power grid electrical parameter dataset and the information-symmetrical power grid node adjacency relationship dataset into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain the first fault prediction result of the target power distribution network; The analysis module 204 is configured to analyze the first fault prediction result to obtain the first correlation information between each fault node corresponding to the first fault prediction result; The dimension reduction module 205 is configured to perform dimension reduction processing on the first power grid electrical parameter dataset to obtain a second power grid electrical parameter dataset; The second fault prediction model module 206 is configured to input the correlation information and the second power grid electrical parameter dataset into a second fault prediction model constructed by a graph neural network algorithm for processing to obtain second correlation information; ​The third fault prediction model module 207 is configured to input the first fault prediction result, the second power grid electrical parameter dataset and the second correlation information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target distribution network. The analysis module 208 is configured to analyze the second fault prediction result to generate a fault location log of the target distribution network.

[0044] The distribution network fault location device 300 can include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art can understand that the schematic diagram is only an example of the distribution network fault location processing device, and does not constitute a limitation on the distribution network fault location processing device 300, and can include more or fewer components than the diagram, or combine certain components, or different components, for example, the distribution network fault location device 300 can also include an input / output device, a network access device, a bus, etc.

[0045] The processor 301 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The processor 301 is a control center of the distribution network fault location device 300, and connects various parts of the entire distribution network fault location device 300 through various interfaces and lines.

[0046] The memory 302 can be used to store the computer programs and / or modules, and the processor 301 realizes various functions of the power distribution network fault location device 300 by running or executing the computer programs and / or modules stored in the memory 302, and calling the data stored in the memory 302. The memory 302 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), and the like. In addition, the memory 302 can include a high-speed random access memory, and can also include a nonvolatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0047] The modules integrated in the power distribution network fault location device 300 can be stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, an executable file, or some intermediate form, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0048] It can be understood by those of ordinary skill in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium, and the program can include the processes of the above-mentioned various method embodiments when executed.

[0049] Correspondingly, an embodiment of the present application provides a computer readable storage medium comprising a stored computer program, wherein the computer readable storage medium controls a device in which the computer readable storage medium is located to perform steps in the power distribution network fault locating method of the above-mentioned embodiment, for example Figure 1 steps S1-S8 described in the above-mentioned embodiment.

[0050] Another embodiment of the present application provides a power distribution network fault locating method, device and locating result of a storage medium. The embodiment uses locating accuracy and one-hop locating accuracy as performance indicators. One-hop locating accuracy is an extension of locating accuracy, which considers neighborhood information of the fault location. The performance indicators of the model are represented by the average of the locating accuracy and the one-hop locating accuracy of all nodes.

[0051] In the embodiment, since fault data of an actual power distribution network is difficult to obtain, the embodiment uses OpenDSS simulation software to simulate and generate fault data of the power distribution network.

[0052] The embodiment generates data based on the framework proposed by the IEEE123 node test feeder system. The graph structure of the IEEE123 node test system is shown in FIG. 1. Figure 5 Figure 5 FIG. 1 shows a graph structure diagram of the IEEE123 node test system in one embodiment of the present application, wherein 1-114 represent a first type of node of the power distribution network, 135, 150, 152, 160, 197, 250, 300, 350 and 450 represent a second type of node (the type is mainly determined by the installation location, model, etc. of the node, which is not specifically limited in the embodiment of the present application), 9r, 25r, 150r and 160r represent voltage regulators, 61s represents a transformer, and 451 represents a switch. Since there is a problem of low observation rate of fault nodes of the power distribution network, the embodiment selects 21 nodes as observation nodes, and the measurement data of the remaining nodes are all 0. Among them, 9 pairs of nodes are connected through switches (-s) or voltage regulators (-r) at the same location. Therefore, there are a total of 119 possible fault locations.

[0053] ​For fault samples, a value in [0.05, 20] is randomly selected as the fault impedance, and fault data simulation is generated for all three-phase nodes (68) in the IEEE123 system, with 0 and 1 indicating whether the node is faulty, 0 indicating no fault, and 1 indicating fault; the fault classification label includes: single-phase-to-ground short-circuit fault (sp), two-phase-to-ground short-circuit fault (ppg), two-phase-to-phase short-circuit fault (pp), three-phase-to-phase short-circuit fault (p3), three-phase-to-ground short-circuit fault (pg3), and no fault (nf), wherein the number of samples for each type of label is 120 samples. For fault data, a total of 40800 samples are generated; for non-fault data, a total of 8160 samples are simulated.

[0054] The 48960 samples including fault samples and non-fault samples are uniformly divided into a training set, a validation set and a test set according to a ratio of 6:1:1.

[0055] For data samples, the data samples are standardized by subtracting the mean value and scaling the standard deviation of all data sets.

[0056] The graph neural network model part sets three convolutional layers and one fully connected layer; wherein the number of hidden layers is 3, and the number of aggregated neighbor information is set considering the sparsity of power grid infrastructure ; in the GCN after the graph learning layer, two convolutional layers are included, the number of neurons is , and one fully connected layer is included, and , i.e. the first similar neighbors of each node have non-zero values. Training is performed on a personal computer equipped with an Intel Core i9-13900, RAM of 56GHz, 48GB and NVIDIA RTX 4080 SUPER, the learning framework uses Pytorch, and various parameter settings in the training process are shown in Table 1.

[0057] Table 1 Training parameter settings The performance of the embodiment compared with the current existing model GraphSAGE is shown in Table 2, and it can be seen that the fault positioning accuracy of the embodiment on the IEEE123 node feeder system is 93.26%, the one-hop positioning accuracy is 97.76%, compared with the GraphSAGE model, the accuracy is increased by 4.06%, and the 1-hop accuracy is increased by 1.16%.

[0058] Table 2 Performance comparison of the model of the embodiment and other models The above experimental results show that the positioning accuracy of the embodiment is higher than that of other existing power distribution network fault positioning models, verifying the effectiveness of the embodiment.

[0059] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it should not be understood as limiting the scope of the patent of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

[0060] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following: (1) By combining the information symmetric power grid node adjacency relationship and electrical parameter data, and using the multi-layer topology perception ability of graph neural network and graph convolution network, the topology features and dynamic operation state of the power grid can be more comprehensively captured, thereby significantly improving the accuracy of fault location prediction and reducing misjudgment and omission.

[0061] (2) A phased processing strategy is adopted, first the possible fault area is preliminarily screened through the graph neural network, and then the multi-layer topology perception network is used to refine and optimize combined with dimension reduction data and associated data, so that the model can adapt to power distribution networks of different scales and structures, and has stronger robustness to noise data and abnormal values.

[0062] (3) Through information symmetric data processing and dimension reduction technology, the redundancy of input data is reduced, combined with the local perception characteristics of graph convolution network, large-scale power grid node data can be efficiently processed, and the model inference speed is accelerated. At the same time, the phased output of fault probability results provides an intermediate explanation of the fault positioning process, which helps operation and maintenance personnel to understand the model decision basis.

Claims

1. A method for fault location in a power distribution network, characterized by, The method comprises the following steps: obtaining node topology information of a target power distribution network to obtain an information-symmetrical power grid node adjacency relationship dataset; obtaining operation information of a selected target node to obtain a first power grid electrical parameter dataset; inputting the first power grid electrical parameter dataset and the information-symmetrical power grid node adjacency relationship dataset into a first fault prediction model constructed by a graph neural network algorithm for processing to obtain a first fault prediction result of the target power distribution network; analyzing the first fault prediction result to obtain first correlation information between each fault node corresponding to the first fault prediction result; performing dimensionality reduction processing on the first power grid electrical parameter dataset to obtain a second power grid electrical parameter dataset; inputting the correlation information and the second power grid electrical parameter dataset into a second fault prediction model constructed by a graph neural network algorithm for processing to obtain second correlation information; inputting the first fault prediction result, the second power grid electrical parameter dataset and the second correlation information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target power distribution network; analyzing the second fault prediction result to generate a fault location log of the target power distribution network.

2. The power distribution network fault location method of claim 1, wherein, The method comprises the following steps: analyzing the node topology information of the target power distribution network, calculating the shortest path between nodes to obtain shortest path information; based on the node topology information and the shortest path information of the target power distribution network, obtaining a power grid node adjacency relationship dataset; using a maximum value transformation function to transform the power grid node adjacency relationship dataset into an information-symmetrical power grid node adjacency relationship dataset.

3. The power distribution network fault location method of claim 1, wherein, The method comprises the following steps: performing aggregation operations on the first power grid electrical parameter dataset and the information-symmetrical power grid node adjacency relationship dataset to obtain network node embedding of the first fault prediction model constructed by the graph neural network algorithm; processing the network node embedding to obtain hidden features of nodes; using two fully connected layers to convert the aggregated hidden features to obtain the first fault prediction result of the target power distribution network; inputting into a fault probability preliminary prediction model using two fully connected layers and a cross-entropy loss function for layer-by-layer aggregation and probability conversion processing to obtain the first fault prediction result of the target power distribution network.

4. The power distribution network fault location method of claim 1, wherein, The first fault prediction result at least includes each first fault node and a fault probability corresponding to each first fault node; the method comprises the following steps: based on the fault probability comparison result, screening each first fault node to obtain a second fault node; determining a fault event corresponding to each second fault node; extracting event features of each of the fault events; obtaining the first correlation information according to the vector cosine distance of each of the event features.

5. The power distribution network fault location method of claim 1, wherein, The processing of inputting the correlation information and the second power grid electrical parameter data set into the second fault prediction model constructed by the graph neural network algorithm respectively to obtain the second correlation information comprises: inputting the second power grid electrical parameter data set into the second fault prediction model constructed by the graph neural network algorithm to obtain the graph learning weight of the second power grid electrical parameter data set; multiplying the graph learning weight and the correlation information to obtain the second correlation information.

6. The power distribution network fault location method of claim 1, wherein, The inputting of the first fault prediction result, the second power grid electrical parameter data set and the second correlation information into the third fault prediction model constructed by the graph convolution network algorithm to obtain the second fault prediction result of the target distribution network comprises: inputting the first fault prediction result, the second power grid electrical parameter data set and the second correlation information into the graph convolution layer of the third fault prediction model for layer-by-layer convolution to obtain a convolution result; inputting the convolution result into a full connection layer to obtain a full connection output result; converting the full connection output result to obtain a second fault location prediction probability result.

7. The power distribution network fault location method of claim 1, wherein, The analysis of the second fault prediction result to generate the fault positioning log of the target distribution network comprises: analyzing the second fault prediction result to obtain the probability of each node of the second fault prediction result occurring the fault event; screening the probability information to determine fault positioning; generating the fault positioning log of the target distribution network based on the fault positioning.

8. A power distribution network fault location apparatus characterized by, Comprise: a topology information acquisition module, configured to acquire node topology information of a target distribution network to obtain an information-symmetrical power grid node adjacency relationship data set; an operation information acquisition module, configured to acquire operation information of a selected target node to obtain a first power grid electrical parameter data set; a first fault prediction model module, configured to input the first power grid electrical parameter data set and the information-symmetrical power grid node adjacency relationship data set into a first fault prediction model constructed by a graph neural network algorithm respectively to process to obtain a first fault prediction result of the target distribution network; an analysis module, configured to analyze the first fault prediction result to obtain first correlation information between each fault node corresponding to the first fault prediction result; a dimension reduction module, configured to perform dimension reduction processing on the first power grid electrical parameter data set to obtain a second power grid electrical parameter data set; a second fault prediction model module, configured to input the correlation information and the second power grid electrical parameter data set into a second fault prediction model constructed by a graph neural network algorithm respectively to process to obtain second correlation information; a third fault prediction model module, configured to input the first fault prediction result, the second power grid electrical parameter data set and the second correlation information into a third fault prediction model constructed by a graph convolution network algorithm to obtain a second fault prediction result of the target distribution network; an analysis module, configured to analyze the second fault prediction result to generate a fault positioning log of the target distribution network.

9. A power distribution network fault location apparatus characterized by, The computer readable storage medium stores a computer program, and a device where the computer readable storage medium is located executes the computer program to realize the power distribution network fault locating method as claimed in any one of claims 1 to 7.

10. A computer readable storage medium characterized by The computer readable storage medium stores a computer program, and a device where the computer readable storage medium is located executes the computer program to realize the power distribution network fault locating method as claimed in any one of claims 1 to 7.

Images