Distribution line fault positioning method based on power frequency and traveling wave data

Abstract

The embodiment of the invention discloses a distribution line fault positioning method based on power frequency and traveling wave data, and the method comprises the steps: obtaining historical traveling wave data when a distribution line fault occurs, carrying out the preprocessing and feature extraction of the historical traveling wave data, and obtaining graph structure data; on the basis of the graph structure data, node feature coding is carried out by using a VGAE model, and node low-dimensional features are obtained; based on the node low-dimensional features, node information is updated by using a GraphSAGE model to obtain updated node feature vectors, and based on the updated node feature vectors, the distribution lines are classified, and faulted lines are identified; the feature information of the nodes in the power distribution network can be extracted and represented more accurately, and the accuracy of fault positioning is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault identification, and in particular to a distribution line fault locating method based on power frequency and traveling wave data. Background Art

[0002] The power system is an indispensable infrastructure in modern society and is crucial to supporting daily life and industrial production. Among them, the distribution network directly serves end users and is responsible for distributing electricity from the transmission network to various power consumption points. With the development of society and the advancement of technology, the distribution network faces the challenges of increasingly complex structures and diverse operating conditions. In particular, with the increase in the access of distributed energy, the randomness and uncertainty of the distribution network have increased, which has brought new challenges to fault location.

[0003] Traditional distribution network fault location technologies mainly include impedance method, signal injection method and traveling wave method. The impedance method measures the voltage and current in the circuit at the time of the fault and calculates the fault location in combination with the line parameters, but this method assumes idealized line conditions and may not be accurate enough in actual situations. The traveling wave method relies on the time difference between the two ends of the traveling wave signal generated at the moment of the fault to locate the fault. Although this method is relatively mature and widely used, its positioning accuracy is sometimes difficult to guarantee due to the need for a highly synchronized time reference and the interference of wave speed changes and network topology uncertainty, and further optimization is needed. The signal injection method identifies the fault location by injecting a signal of a specific frequency or mode into the line and tracing its propagation path. Although this method can achieve online monitoring, the high operation and maintenance costs limit its popularization and application.

[0004] It can be seen that there is an urgent need for a distribution network fault location method that is highly accurate and easy to popularize. Summary of the invention

[0005] The main purpose of the present invention is to provide a distribution line fault location method based on power frequency and traveling wave data, which can improve the accuracy of fault location.

[0006] To achieve the above objectives, the present application provides, in a first aspect, a method for locating a distribution line fault based on power frequency and traveling wave data, the method comprising:

[0007] Acquire historical traveling wave data when a distribution line fault occurs, preprocess and extract features from the historical traveling wave data, and obtain graph structure data;

[0008] Based on the graph structure data, a VGAE model is used to encode node features to obtain low-dimensional features of nodes;

[0009] Based on the low-dimensional features of the nodes, the node information is updated using the GraphSAGE model to obtain an updated node feature vector, and based on the updated node feature vector, the distribution lines are classified to identify the faulty lines.

[0010] A second aspect of the present application provides a distribution line fault location device, comprising:

[0011] A data processing module, used to obtain historical traveling wave data when a distribution line fault occurs, preprocess and extract features from the historical traveling wave data, and obtain graph structure data;

[0012] An encoding module, used to encode node features based on the graph structure data using a VGAE model to obtain low-dimensional features of the nodes;

[0013] The identification module is used to update the node information based on the low-dimensional features of the node using the GraphSAGE model to obtain an updated node feature vector, and classify the distribution lines based on the updated node feature vector to identify the faulty lines.

[0014] A third aspect of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the first aspect and any possible implementation thereof.

[0015] A fourth aspect of the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes each step of the method described in the first aspect.

[0016] The present application provides a distribution line fault location method based on power frequency and traveling wave data, by obtaining historical traveling wave data when a distribution line fault occurs, preprocessing and feature extraction are performed on the historical traveling wave data to obtain graph structure data; based on the graph structure data, node feature encoding is performed using a VGAE model to obtain low-dimensional features of the nodes; based on the low-dimensional features of the nodes, node information is updated using a GraphSAGE model to obtain updated node feature vectors; based on the updated node feature vectors, the distribution lines are classified to identify the faulty lines; the feature information of the nodes in the distribution network can be more accurately extracted and represented, wherein the VGAE model enhances the resistance to noise by learning the potential representation of the nodes and provides deep-level abstract features, and the GraphSAGE model fully considers the influence of surrounding lines on the target line through neighbor sampling and aggregation operations, thereby improving the accuracy of fault location. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] in:

[0019] Figure 1 A schematic diagram of a flow chart of a method for locating a distribution line fault based on power frequency and traveling wave data provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of a topological structure of a distribution network provided in an embodiment of the present application;

[0021] Figure 3 A schematic diagram of the structure of a distribution line fault location device provided in an embodiment of the present application;

[0022] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with 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 of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0024] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0025] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0026] The graph structured data involved in the embodiments of the present application is a data structure used to represent the relationship between entities (nodes). In the graph structure, the data consists of nodes (or vertices) and edges (or links) connecting these nodes.

[0027] The graph convolutional network (GCN) involved in the embodiments of the present application is a deep learning model for processing graph structure data. It generalizes the convolution operation to the graph structure, so that the model can learn the representation of nodes, which can be used for various tasks, such as node classification, graph classification, link prediction, etc.

[0028] The core idea of ​​GCN is to use the adjacency matrix of the graph to capture the connection relationship between nodes and update the feature representation of the nodes through the neural network layer. In GCN, the feature update of each node depends not only on itself, but also on the features of its neighboring nodes. This feature propagation mechanism based on the adjacency matrix enables GCN to effectively capture the topological information of the graph structure.

[0029] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0030] See also Figure 1 , is a flow chart of a method for locating a distribution line fault based on power frequency and traveling wave data provided in an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0031] 101. Obtain historical traveling wave data when a distribution line fault occurs, preprocess and extract features from the historical traveling wave data, and obtain graph structure data.

[0032] The execution subject of the method in the embodiment of the present application may be a distribution line fault locating device, which may be implemented using a terminal device, such as a server or a computer, in actual applications.

[0033] The above historical traveling wave data can be collected and stored by a detection device.

[0034] Traveling wave detection is a technology used for fault location in power systems. Its basic principle is to use the voltage and current traveling waves generated on the power line when a fault occurs to perform fault distance measurement and location.

[0035] In an optional implementation manner, the above-mentioned obtaining of historical traveling wave data when a distribution line fault occurs includes:

[0036] By setting a traveling wave detection device at both ends of the above-mentioned distribution line, the traveling wave signal when the above-mentioned distribution line fault occurs is detected, and the detection signal is transmitted to the signal processing unit through a wireless sensor and stored in a database;

[0037] Read the above historical traveling wave data from the above database;

[0038] The above-mentioned historical traveling wave data is preprocessed and feature extracted to obtain graph structure data, including:

[0039] A filter is used to remove background noise from the above historical traveling wave data, and normalization is performed to integrate the data into a data set, and the data set is converted into the above graph structure data.

[0040] Specifically, a traveling wave detection device can be set at both ends of the distribution network line to detect the traveling wave signal when a distribution line fault occurs, and the detection signal can be transmitted to the signal processing unit through a wireless sensor as a database;

[0041] In the embodiments of the present application, data preprocessing can be performed: historical traveling wave data is read from the database, background noise in the traveling wave data is removed using a filter, all data are normalized to a uniform range for model training, the traveling wave data is integrated into a unified data set, and the preprocessed data is converted into a format suitable for a graph convolutional network (GCN).

[0042] In an optional implementation, the above method further includes:

[0043] Generate an undirected unweighted isomorphic graph G = (V, E) based on the distribution network, where the node V represents the bus and the edge E represents the feeder branch;

[0044] The above-mentioned data set is converted into the above-mentioned graph structure data, including:

[0045] The topological structure of the above power distribution network is represented by an adjacency matrix;

[0046] Generate node attribute characteristics of each node in the above-mentioned distribution network. The characteristic vector of each of the above-mentioned nodes includes but is not limited to one or more of three-phase voltage, three-phase current, three-phase traveling wave current, effective value, phase angle, active power and reactive power, as well as degree centrality, eigenvector centrality, betweenness centrality and proximity centrality.

[0047] First, in the embodiment of the present application, the power distribution network can be abstracted as an undirected unweighted isomorphic graph, in which nodes represent buses and edges represent feeder branches. The structural information of the graph is represented in the form of an adjacency matrix A, and the node attribute characteristics are described by a matrix X. These two matrices are provided to the model as input.

[0048] 102. Based on the above graph structure data, the VGAE model is used to encode node features to obtain low-dimensional features of the nodes.

[0049] The variational graph auto-encoder (VGAE) mentioned in the embodiments of the present application is an unsupervised learning model that combines the ideas of graph autoencoder (GAE) and variational autoencoder (VAE) for node embedding learning of graph structured data.

[0050] First, VGAE is used in the embodiments of the present application to deeply explore the underlying features of each node in the distribution network (such as transformers, circuit breakers and other equipment). VAGE is a generative model based on graph neural networks, which can learn a probability distribution representing the potential state of each node. In this way, VGAE not only captures the characteristics of the node itself, but also takes into account its position in the network and its relationship with other nodes. More importantly, because VGAE uses variational inference technology, it can naturally handle uncertainty in the data and has strong noise resistance. This means that even in the face of data containing a large amount of random disturbances or measurement errors, the model can still provide relatively stable and reliable node representation.

[0051] In an optional implementation, the above step 102 includes:

[0052] Input the adjacency matrix and node attribute features in the above graph structure data;

[0053] The VGAE model based on graph convolutional network is used to map the above node attribute features to independent Gaussian distribution;

[0054] Perform reparameterization and draw samples from the above Gaussian distribution to obtain a low-dimensional feature vector for each node.

[0055] Specifically, using VGAE to encode node features mainly includes:

[0056] Input processing: take the adjacency matrix A and node attribute feature X as input;

[0057] Feature mapping: The VGAE model based on graph convolutional network is used to map the node attribute feature X to an independent Gaussian distribution N(μi,σi 2 ), where μi and σi2 are the mean and variance of each node feature respectively;

[0058] Reparameterization: In order to make the sampling operation participate in the gradient descent process, the reparameterization trick is used to extract samples from the above Gaussian distribution to obtain a low-dimensional abstract feature vector for each node.

[0059] In this way, the encoder output not only contains more useful deep abstract features, but also enhances the resistance to noise interference.

[0060] Specifically, in this model, buses can be divided into two categories: system buses and load buses; load buses are directly connected to load devices;

[0061] Node (V): Each node represents a bus.

[0062] Edges (E): Each edge represents a feeder branch between two buses, and these edges are undirected and unweighted.

[0063] Input components: When building a graph neural network model, the input data consists of two parts:

[0064] Topological information: The topological structure of the distribution network is represented by the adjacency matrix A. If the graph G has N nodes, the dimension of the adjacency matrix A is N×N, which is defined as follows:

[0065]

[0066] Here, A[i][j] indicates whether there is an edge between node i and node j.

[0067] The attribute characteristics of each node can include the following information:

[0068] The amplitude and phase angle of the three-phase voltage: V 1 ,θ V1 , V 2 ,θ V2 , V 3 ,θ V3 ;

[0069] The amplitude and phase angle of the three-phase current: I 1 ,θ I1 , I2 ,θ I2 , I 3 ,θ I3 ;

[0070] Active power P and reactive power Q;

[0071] In addition, four centrality features based on graph theory calculations can also be included in node features: degree centrality (DC), eigenvector centrality (EC), betweenness centrality (BC), and closeness centrality (CC).

[0072] Therefore, the feature vector of each node can be expressed as:

[0073]

[0074] The data samples of the entire distribution network can be represented as a matrix Where N is the number of nodes and each row corresponds to the feature vector of a node.

[0075] Specifically, the distribution network is abstracted as graph data G = (V, E), where V represents the node set and E represents the edge set. The number of nodes is represented by N = |V|. The adjacency matrix of the graph G is defined as A, and the degree matrix is ​​defined as D. The dimension of the node input feature matrix X is N × Fin, where Fin is the input feature dimension of each node.

[0076] Optionally, further introduce hidden variables z i , the matrix Z consisting of the hidden representation of each node has dimension N×F, where F represents the feature dimension of the hidden representation of the node.

[0077] In a specific implementation, the variational graph autoencoder (VGAE) model structure is:

[0078] The encoder is a two-layer graph convolutional neural network (GCN) that generates the hidden representation Z of the node;

[0079] The decoder reconstructs the graph by calculating the probability that there is an edge between any two points in the sample variable. Specifically, the decoder calculates the probability that two nodes z i and z j The connection probability between:

[0080]

[0081] in,

[0082] The loss function of the VGAE model includes reconstruction error and KL divergence terms:

[0083]

[0084] Where p(Z) is the standard normal distribution:

[0085]

[0086] In order to simplify the computational complexity of the model and increase the training speed, the matrix inner product in the original decoder is changed to a single-layer fully connected network. The calculation formula of the reconfigured adjacency matrix is ​​as follows:

[0087]

[0088] Among them, W and B are the weight matrix and bias term of the fully connected network respectively.

[0089] In one embodiment, the training process may include:

[0090] The parameters of the VGAE model are updated using the Mini-batch Gradient Descent optimizer. The dataset is divided into batches (batch sizes), and a batch of data is input for training each time. The model parameters are updated by gradient descent based on the average value of the loss function calculated for each batch of samples, and the data in the entire training set is passed as an epoch. After training, the VGAE model parameters are saved. Through the encoder of the VGAE model, abstract feature representations of each node in the distribution network can be obtained, which can be used as input for downstream tasks (such as distribution network fault location).

[0091] In this way, the VGAE model can effectively extract deep features of distribution network nodes and provide more powerful representation capabilities in subsequent tasks.

[0092] 103. Based on the low-dimensional features of the nodes, the node information is updated using the GraphSAGE model to obtain an updated node feature vector. Based on the updated node feature vector, the distribution lines are classified to identify the faulty lines.

[0093] GraphSAGE (Graph Sample and Aggregation) in the embodiment of the present application is a graph neural network model for graph node embedding learning. Its core idea is to generate low-dimensional vector representations of nodes by sampling and aggregating feature information of neighboring nodes. These representations can be used for various machine learning tasks, such as node classification, clustering, and link prediction.

[0094] In an optional embodiment, the above GraphSAGE model adopts a pooling aggregation function;

[0095] Based on the low-dimensional features of the nodes, the GraphSAGE model is used to update the node information and obtain the updated node feature vector, which may include:

[0096] Based on the low-dimensional feature vector of the node, the GraphSAGE model is used to perform sampling and information aggregation of neighbor nodes, and the information of each node is gradually updated to obtain the updated node feature vector.

[0097] Specifically, using GraphSAGE to update node information mainly includes:

[0098] Sampling and aggregation: Based on the low-dimensional feature Z of the node obtained in the previous step, the GraphSAGE model is used to perform sampling and information aggregation of neighboring nodes, fully considering the impact of surrounding lines on the target line;

[0099] Node information update: Through multi-layer aggregation operations, the information of each node is gradually updated so that the node representation more comprehensively reflects its position in the network and its relationship with other nodes.

[0100] In an optional implementation, the above-mentioned classification of the distribution lines based on the above-mentioned updated node feature vectors to identify the faulty lines includes:

[0101] The updated node feature vector is processed through a fully connected layer to extract higher-level features. The output of the fully connected layer is used to classify the distribution lines and identify the faulty lines.

[0102] For the fault location output, consider using a fully connected layer for processing. Specifically, the output of the GraphSAGE model can be further processed through a fully connected neural network layer to extract higher-level features. Finally, through the output of the fully connected layer network, the distribution line is classified, thereby accurately identifying the faulty line.

[0103] Further optionally, in the above-mentioned process of sampling and information aggregation of neighbor nodes, the node matrix is ​​set to have three layers:

[0104] The feature vector of a node in the second layer is aggregated from the feature vectors of its first-order neighbors;

[0105] In the first layer, the number of sampled neighbors is set to 3, and the neighbor set of the node is obtained by random sampling without replacement;

[0106] At level 0, the feature vector of each node is aggregated from the feature vectors of its first-order neighbors; and

[0107] The number of sampled neighbors is set differently for each layer. As the number of layers increases, the range of information contained in the node feature vector expands.

[0108] In the embodiment of the present application, a variational graph autoencoder (VGAE) is first used to extract the hidden layer abstract features of the bus nodes in the distribution network. Then, the adjacency matrix A of the original graph is processed using the GraphSAGE model to realize the transmission and aggregation of neighbor information.

[0109] Specifically, the sampling and aggregation process of GraphSAGE can be as follows:

[0110] The node matrix is ​​set up with three layers. The feature vector of the node in the second layer is aggregated from the feature vectors of its first-order neighbors. In the first layer, the number of sampled neighbors is set to 3, and the node's neighbor set is obtained by random sampling without replacement. Continuing down, in the 0th layer, the feature vector of each node is obtained by aggregating its first-order neighbors, and the number of sampled neighbors in each layer is set differently. As the number of layers increases, the range of information contained in the node feature vector expands, but too many layers may lead to the loss of local characteristics. Therefore, it is necessary to choose the number of layers reasonably.

[0111] Selection of aggregate function:

[0112] GraphSAGE provides a variety of aggregation methods, including mean aggregation, pooling aggregation, and LSTM aggregation. Considering the model complexity and performance, the embodiment of the present application can select the pooling aggregation function. Pooling aggregation can maintain a relatively low number of parameters while retaining deep information.

[0113] Feature Updates:

[0114] For the update of each layer of node features, the following formula is used:

[0115]

[0116] Among them, =CONCAT() represents vector concatenation, and σ is the ReLU activation function.

[0117] Fault location output:

[0118] The final output of the GraphSAGE model is the updated node feature vector. In order to locate the fault, these feature vectors are input into a fully connected layer, and its output dimension is set to the number of distribution network lines. The fault probability of each line is calculated by the softmax function:

[0119]

[0120] Loss function and training:

[0121] Fault localization is a multi-classification task, so cross entropy is used as the loss function:

[0122]

[0123] Among them, y i is the predicted probability distribution, is the actual failure probability distribution.

[0124] During the back propagation process, the parameters of GraphSAGE and the fully connected layer can be adjusted according to the loss function so that the prediction results of the model gradually approach the true label. After the training is completed, the model parameters are saved for subsequent calls to perform the distribution network line fault location task.

[0125] The embodiment of the present application mainly provides a distribution network fault location method based on a variational graph autoencoder combined with a graph sampling and aggregation framework. First, VGAE is used to deeply explore the underlying features of each node in the distribution network (such as transformers, circuit breakers and other equipment). VAGE is a generative model based on a graph neural network, which can learn a probability distribution representing the potential state of each node. In this way, VGAE not only captures the characteristics of the node itself, but also considers its position in the network and its relationship with other nodes. More importantly, since VGAE uses variational inference technology, it can naturally handle uncertainty in the data and has strong noise resistance. This means that even in the face of data containing a large amount of random perturbations or measurement errors, the model can still provide a relatively stable and reliable node representation. Then, the neighbor information is aggregated through the GraphSAGE model, and these high-quality node embeddings generated by VGAE are passed as input to the GraphSAGE model. GraphSAGE is an algorithm specifically designed to process large-scale graph data. It updates the representation of each node by sampling the local neighborhood and performing multi-layer aggregation operations. In this process, GraphSAGE fully considers the impact of surrounding lines on the target line, that is, the state of each node is not only affected by its own attributes, but also by the states of other nodes directly and indirectly connected to it. In this way, we can understand the mechanism of fault occurrence more broadly, thereby improving the accuracy and robustness of fault location.

[0126] The following is an analysis and study of specific cases to introduce the method in the embodiments of the present application.

[0127] In order to verify the superiority of the model combining VGAE and GraphSAGE in distributed network fault location, it is compared with several other models in the embodiments of the present application, including the convolutional neural network (CNN) in Euclidean space, the fully connected network (FC) in non-Euclidean space, the graph convolutional network (GCN) and the graph attention network (GAT);

[0128] The simulation node network is simulated by traveling wave data through ATP-EMTP, and the power frequency data is simulated by OPenDSS. All models are trained and tested using the same data set.

[0129] The simulation test is carried out using IEEE123 node network. Figure 2 A schematic diagram of the topological structure of a distribution network provided in an embodiment of the present application. Each dot represents a node, which may be a substation, a distribution transformer, a switchgear, or a user terminal, and the lines connecting these dots represent power lines. The nodes and lines together constitute the topological structure of the distribution network. The node identification includes a number on the node, which is the node number, which helps to identify and refer to a specific node.

[0130] Through ATP-EMTP and OpenDSS, fault simulation is performed based on the above distribution network to measure the positioning accuracy and positioning time of different models for traveling wave data and power frequency data, as shown in Table 1.

[0131] Model Positioning accuracy (%) Positioning time(s) CNN 85.3 0.8 FC 82.1 1.2 GCN 90.7 0.6 GAT 92.4 0.5 VGAE-GraphSAGE 95.1 0.7

[0132] Table 1

[0133] Graph neural networks (such as GCN, GAT and VGAE-GraphSAGE) show significant advantages in processing graph structure data of distribution networks. Their positioning accuracy is generally higher than that of traditional CNN and FC models. In terms of positioning speed, CNN and FC models take the longest time because they require more layers and parameters to fit complex relationships. In contrast, GCN and GAT models reduce the number of features and improve positioning efficiency through convolution and attention mechanisms. Although the VGAE-GraphSAGE model proposed in the embodiment of the present application is slightly slower than GAT in positioning speed, it achieves higher positioning accuracy and the positioning time is still within an acceptable range.

[0134] Another complete specific embodiment is provided below:

[0135] Traveling wave detection devices are installed at both ends of the distribution network, and the arrival time of the traveling waves at both ends is calculated to estimate the location of the fault point. The detected data is transmitted to the central processing unit for storage through the traveling wave detection device, and the central processing unit imports the traveling wave signal into the model through data preprocessing. Specifically:

[0136] The system includes a traveling wave detection device, a data transmission module, a central processing unit and a VGAE-GraphSAGE model; the traveling wave data is recorded by the traveling wave detection device and transmitted to the storage module as a database for data training to optimize the fault location result.

[0137] The specific steps involve:

[0138] Traveling wave detection: When a fault occurs in the distribution line, the traveling wave detection device detects the traveling wave signal at the A end and the B end respectively.

[0139] The time tA and tB when the two devices detect the traveling wave signal are recorded, and other relevant electrical parameters such as three-phase voltage, three-phase current, three-phase traveling wave current, etc. are recorded. The detected time tA and tB, as well as the traveling wave current and the traveling wave voltage are sent to the central processing unit through a data transmission module (such as optical fiber communication or wireless communication). The central processing unit stores the received data in a database for subsequent processing and analysis. The database uses a relational database (such as MySQL) to remove outliers and noise to ensure the accuracy of the data. A bandpass filter is used to remove background noise in the traveling wave data. Wavelet transform is used for further denoising, and key features such as arrival time, amplitude, polarity, etc. of traveling waves are extracted. All data are normalized to a unified range (e.g., 0-1 or -1 to 1) to facilitate model training. Key features of each node are extracted, including three-phase voltage, three-phase current, three-phase traveling wave current, effective value, phase angle, active power and reactive power, etc. These features are integrated into a unified data set. The feature vector of each node includes the above information, and the distribution network is abstracted into an undirected unweighted isomorphic graph G = (V, E), where node V represents the bus and edge E represents the feeder branch.

[0140] Input processing: Take the adjacency matrix A and node attribute feature X as input.

[0141] Feature mapping: The VGAE model based on graph convolutional network (GCN) is used to map the node attribute feature X to an independent Gaussian distribution N(μi,σi2), where μi and σi2 are the mean and variance of each node feature, respectively.

[0142] Reparameterization: In order to enable the sampling operation to participate in the gradient descent process, the reparameterization trick is used to extract samples from the above Gaussian distribution to obtain a low-dimensional abstract feature vector Z for each node.

[0143] Loss function: The reconstruction error is usually calculated by calculating the predicted adjacency matrix The difference between the actual adjacency matrix A and the original adjacency matrix A is obtained. For the binary adjacency matrix, the binary cross entropy (BCE) is used as the loss function:

[0144]

[0145] Where: N is the number of nodes in the graph;

[0146] A ij is an element in the true adjacency matrix;

[0147] is the corresponding element in the adjacency matrix decoded by the VGAE model.

[0148] The KL divergence measures the Gaussian distribution q(z) of each node feature vector generated from the model. i |x i ) and the standard normal prior p(z i ), for each node i, its corresponding latent variable z i The posterior distribution of is assumed to be a multivariate Gaussian distribution with mean μ i and the diagonal covariance matrix Therefore, the KL divergence formula is:

[0149]

[0150] where μ i is the mean vector output by the encoder;

[0151] is the square of the variance vector output by the encoder;

[0152] I is the identity matrix.

[0153] Use GraphSAGE to update node information:

[0154] Sampling and aggregation: Based on the low-dimensional feature Z of the node obtained in the first step, the GraphSAGE model is used to perform sampling and information aggregation of neighboring nodes, fully considering the impact of surrounding lines on the target line. Through multi-layer aggregation operations, the information of each node is gradually updated so that the node representation more comprehensively reflects its position in the network and its relationship with other nodes. The output of the GraphSAGE model is further processed through a fully connected neural network layer to extract higher-level features. Finally, the classification of distribution lines is realized through the output of the fully connected layer network, so that the faulty lines can be accurately identified.

[0155] After multiple trainings, the model has significantly improved the accuracy of locating the traveling wave signal generated by the fault compared to the calculated fault location point.

[0156] The steps involved can also be found in Figure 1 The specific description of the illustrated embodiments will not be repeated here.

[0157] The embodiment of the present application can more accurately extract and represent the characteristic information of nodes in the distribution network by combining the variational graph autoencoder (VGAE) and the graph sampling and aggregation framework (GraphSAGE). The VGAE model enhances the resistance to noise and provides deep abstract features by learning the potential representation of nodes. The GraphSAGE model fully considers the impact of surrounding lines on the target line through neighbor sampling and aggregation operations, thereby improving the accuracy of fault location. The model can achieve high-precision fault location under different types of fault settings, especially in the case of single-phase grounding faults and distributed power supply access.

[0158] The VGAE-GraphSAGE model proposed in the embodiment of the present application shows remarkable robustness and generalization ability when dealing with distributed power access and load changes. By introducing randomly sampled neighbor nodes, the GraphSAGE model increases the flexibility of the model, enabling it to better adapt to changes in network topology. The VGAE model further enhances the model's resistance to data noise and uncertainty by learning the potential distribution of nodes. Even in a complex distribution network environment, the model can still maintain a high level of fault location accuracy, reducing the possibility of misjudgment and missed detection.

[0159] The embodiments of the present application significantly improve computing efficiency and scalability by optimizing the model structure and training process. The GraphSAGE model reduces the amount of computation through neighbor sampling, so that the model can run efficiently in large-scale distribution networks. The VGAE model solves the problem that the sampling operation cannot participate in gradient descent through reparameterization techniques, further improving the training efficiency of the model. Compared with traditional convolutional neural networks (CNNs) and fully connected networks (FCs), the VGAE-GraphSAGE model significantly reduces computing time and resource consumption while maintaining high accuracy, and is suitable for real-time fault location systems.

[0160] Based on the description of the aforementioned method embodiment, an embodiment of the present application further provides a distribution line fault locating device.

[0161] Figure 3 This is a schematic diagram of the structure of a distribution line fault location device provided in an embodiment of the present application. Figure 3 As shown, the distribution line fault location device 300 includes:

[0162] The data processing module 310 is used to obtain historical traveling wave data when a distribution line fault occurs, preprocess and extract features of the historical traveling wave data, and obtain graph structure data;

[0163] The encoding module 320 is used to encode node features based on the graph structure data using the VGAE model to obtain low-dimensional features of the nodes;

[0164] The identification module 330 is used to update the node information based on the low-dimensional features of the nodes using the GraphSAGE model to obtain an updated node feature vector, and classify the distribution lines based on the updated node feature vector to identify the faulty lines.

[0165] Understandably, this involves Figure 3 The relevant contents of each module in the above method embodiment have been described in detail, and the details can be referred to the contents in the method embodiment; that is, Figure 3 A distribution line fault location device 300 is provided which can perform the following operations: Figure 1 Any steps in the illustrated embodiment will not be described in detail here.

[0166] In one embodiment of the present application, an electronic device is also provided. Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown, the electronic device 400 includes a processor 401 and a memory 402, wherein the memory 402 stores a computer program. When the computer program is executed by the processor 401, the following operations are performed: Figure 1 Any step in the method embodiment shown. The electronic device 400 may also include an input / output device, etc. In a specific implementation, the electronic device may be a terminal device, etc.

[0167] In one embodiment, a computer-readable storage medium is further provided. The computer-readable storage medium stores a computer program. When the computer program is executed by the processor 401, the processor 401 executes any step in the above method embodiment.

[0168] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0169] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0170] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A distribution line fault location method based on power frequency and traveling wave data, characterized in that: The method comprises: Acquire historical traveling wave data when a distribution line fault occurs, perform preprocessing and feature extraction on the historical traveling wave data, and obtain graph structure data; Based on the graph structure data, a VGAE model is used to encode node features to obtain low-dimensional features of nodes; Based on the low-dimensional features of the nodes, the node information is updated using the GraphSAGE model to obtain an updated node feature vector, and based on the updated node feature vector, the distribution lines are classified to identify the faulty lines.

2. The method for locating a distribution line fault based on power frequency and traveling wave data according to claim 1, characterized in that: The method of encoding node features based on the graph structure data using the VGAE model to obtain a low-dimensional feature vector of the node includes: Inputting the adjacency matrix and node attribute features in the graph structure data; A VGAE model based on a graph convolutional network is used to map the node attribute features to independent Gaussian distributions; Reparameterization is performed to extract samples from the Gaussian distribution to obtain a low-dimensional feature vector for each node.

3. The method for locating distribution line faults based on power frequency and traveling wave data according to claim 2 is characterized in that: The GraphSAGE model adopts a pooling aggregation function; The updating of node information based on the low-dimensional features of the node using the GraphSAGE model to obtain an updated node feature vector includes: Based on the low-dimensional feature vector of the node, the GraphSAGE model is used to perform sampling and information aggregation of neighbor nodes, and the information of each node is gradually updated to obtain the updated node feature vector.

4. The method for locating distribution line faults based on power frequency and traveling wave data according to claim 3 is characterized in that: During the sampling and information aggregation process of the neighbor nodes, the node matrix is ​​set up with three layers: The feature vector of a node in the second layer is aggregated from the feature vectors of its first-order neighbors; In the first layer, the number of sampled neighbors is set to 3, and the neighbor set of the node is obtained by random sampling without replacement; At level 0, the feature vector of each node is aggregated from the feature vectors of its first-order neighbors; and, The number of sampled neighbors is set differently for each layer. As the number of layers increases, the range of information contained in the node feature vector expands.

5. The method for locating distribution line faults based on power frequency and traveling wave data according to claim 1, characterized in that: The step of classifying the power distribution lines based on the updated node feature vectors and identifying the faulty lines comprises: The updated node feature vector is processed through a fully connected layer to extract higher-level features, and the distribution line is classified through the output of the fully connected layer to identify the faulty line.

6. The method for locating faults in distribution lines based on power frequency and traveling wave data according to claim 1, characterized in that: The obtaining of historical traveling wave data when a distribution line fault occurs includes: By arranging traveling wave detection devices at both ends of the distribution line, the traveling wave signal when the distribution line fault occurs is detected, and the detection signal is transmitted to the signal processing unit through the wireless sensor and stored in the database; Read the historical traveling wave data from the database; The preprocessing and feature extraction of the historical traveling wave data to obtain graph structure data includes: A filter is used to remove background noise in the historical traveling wave data, and normalization is performed to integrate the historical traveling wave data into a data set, and the data set is converted into the graph structure data.

7. The method for locating distribution line faults based on power frequency and traveling wave data according to claim 6, characterized in that: The method further comprises: Generate an undirected unweighted isomorphic graph G = (V, E) based on the distribution network, where the node V represents the bus and the edge E represents the feeder branch; The converting the data set into the graph structure data comprises: Representing the topological structure of the power distribution network by an adjacency matrix; Generate node attribute characteristics of each node in the distribution network, and the characteristic vector of each node includes but is not limited to one or more of three-phase voltage, three-phase current, three-phase traveling wave current, effective value, phase angle, active power and reactive power, as well as degree centrality, eigenvector centrality, betweenness centrality and proximity centrality.

8. A distribution line fault location device, characterized in that: include: A data processing module, used to obtain historical traveling wave data when a distribution line fault occurs, preprocess and extract features from the historical traveling wave data, and obtain graph structure data; An encoding module, used to encode node features based on the graph structure data using a VGAE model to obtain low-dimensional features of the nodes; The identification module is used to update the node information based on the low-dimensional features of the node using the GraphSAGE model to obtain an updated node feature vector, and classify the distribution lines based on the updated node feature vector to identify the faulty lines.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

Images