A power distribution network fault line selection method considering topology structure change

Abstract

The application relates to the technical field of electric power operation and maintenance, and particularly discloses a power distribution network fault line selection method considering topological structure changes, which comprises the following steps: collecting zero sequence current data and converting the zero sequence current data into graph structure data to provide input information representing the topological structure and current characteristics of the power distribution network for a model; learning the adjacency matrix and node features by using a graph attention network until the model can accurately predict the probability of single-phase ground fault of the power distribution network; setting a threshold based on the node fault probability output by the model to identify and locate a specific fault line; and inputting field zero sequence current data into the trained model to quickly and accurately determine the fault position of the power distribution network. The power distribution network fault line selection method considering topological structure changes can improve the efficiency and accuracy of fault line selection, thereby realizing effective fault monitoring in a wider power distribution network environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power operation and maintenance, and particularly relates to a power distribution network fault line selection method considering topology structure changes. BACKGROUND

[0002] Safe and reliable operation of the power distribution network is an important foundation for realizing the national strategy of developing smart grids and distributed energy. There has been a long-standing problem of fault detection (line selection, positioning) in the power distribution network at home and abroad. Fault detection (line selection, positioning) in the power distribution network is one of the keys to ensuring power distribution safety. However, due to the complex topology structure of the power distribution network, which usually contains multiple branches and nodes according to power demand, high resistance grounding faults such as tree barriers, incomplete lightning arresters, and conductor drops occur frequently on the power distribution line. Long-term operation of the system with high resistance grounding faults may also cause two-phase grounding faults, expanding the fault range and fault nature, and posing a greater risk to personal safety. Therefore, for high resistance grounding faults, accurate detection (line selection, positioning) and arc extinction, overvoltage protection, or even removal of the fault line or section are required.

[0003] Currently, the access of distributed power sources and new lines has made the topology structure of the power distribution network more complex and unstable. Existing fault line selection and positioning methods are no longer suitable for complex network environments containing these factors. Among them, the fault line selection method based on traveling waves is fast in response, but is easily affected by the terrain and line structure, resulting in a decrease in the accuracy of line selection; the fault line selection method based on protection device action information relies on the setting of protection device thresholds, and often causes misoperation, misoperation, and false alarms; the fault line selection method based on artificial intelligence often relies on fixed network structures and pre-set parameters, and this dynamic change brings additional difficulty to fault detection, because the fault characteristics may change due to changes in the network structure, reducing the accuracy and reliability of such methods. In addition, most existing fault detection methods can accurately determine the location of the fault, but they lack comprehensive consideration of the topology structure of the entire power distribution network. These methods cannot fully utilize the structural information of the power distribution network to optimize the fault line selection process, thereby affecting the comprehensiveness and depth of fault line selection.

[0004] To address these challenges, the present application is dedicated to developing a power distribution network fault line selection method considering topology structure changes, which not only can quickly and accurately locate the fault line, but also can comprehensively consider the topology structure information of the power distribution network and adaptively adjust to match the dynamically changing structure of the power distribution network, thereby improving the efficiency and accuracy of fault line selection, and achieving effective fault monitoring in a wider power distribution network environment. SUMMARY

[0005] An object of the present application is to provide a power distribution network fault line selection method considering topology changes, which can not only quickly and accurately locate the fault line, but also comprehensively consider the topology information of the power distribution network and adaptively adjust to match the dynamically changing structure of the power distribution network, thereby improving the efficiency and accuracy of fault line selection, and thus achieving effective fault monitoring in a wider power distribution network environment.

[0006] To achieve the above object, the present application provides a power distribution network fault line selection method considering topology changes, comprising:

[0007] S10: data preprocessing: by collecting zero sequence current data and converting it into graph structure data, input information representing the topology and current characteristics of the power distribution network is provided for the model;

[0008] S20: model training: using a graph attention network to learn the adjacency matrix and node features until the model can accurately predict the probability of single-phase ground fault of the power distribution network;

[0009] S30: fault judgment: based on the node fault probability output by the model, a threshold is set to identify and locate the specific fault line;

[0010] S40: model application: input the field zero sequence current data into the trained model to quickly and accurately judge the fault location of the power distribution network.

[0011] Optionally, the S10 comprises:

[0012] S101: data acquisition: collecting zero sequence current data from the power distribution network field;

[0013] S102: data cleaning: removing noise and outliers from the collected zero sequence current data;

[0014] S103: feature extraction: extracting key features from the cleaned data;

[0015] S104: constructing a graph structure: converting the node and branch information of the power distribution network into graph structure data, where the nodes represent each part of the power distribution network, and the edges represent the connection relationship between the nodes.

[0016] Optionally, the key features extracted from the data include current amplitude and / or phase angle.

[0017] Optionally, the S104 comprises:

[0018] First, the topological connection relationship of the measurement nodes of the power distribution network is obtained, and a node adjacency matrix is made according to formula (1); by measuring the data of the zero sequence current transformer in the field, the required zero sequence current data samples for the experiment are obtained;

[0019] The sample data obtained by field measurement is normalized by formula (4) for each measuring point of the fault current signal to obtain the processed experimental data, which is taken as the node feature data of each node; the nodes on both sides of the fault section of the IEEE18 node network are taken as fault nodes, and the labels of the fault nodes are set to 1 and the labels of the non-fault nodes are set to 0:

[0020] Specifically,

[0021] From the perspective of graph, without considering the internal components and line characteristics, the node network of the power distribution network is abstracted as graph structure data G=(X, A) of nodes and edges; the adjacency matrix A describes the connection relationship of each node of the power distribution network, wherein the row and column numbers of the adjacency matrix correspond to the nodes, as shown in formula (1):

[0022] (1) ;

[0023] The branch zero sequence current is used as the feature vector of the first end node of the branch; in the power system, the zero sequence current is the current generated when the vector sum of the three-phase currents is not zero in the three-phase system; the zero sequence current is usually related to the ground fault; in the actual scene, for the cable line, the zero sequence current transformer is a device specially used to detect the zero sequence current, which is installed on the neutral point of the three-phase power system and can detect the zero sequence current generated due to the fault or unbalanced load; the zero sequence current is collected by using the zero sequence current transformer; and the current measurement of the overhead line usually uses three-phase current transformers, which are respectively installed on each line of the three-phase; by measuring the three-phase currents, the zero sequence current can be calculated; the calculation method is as follows:

[0024] (2) ;

[0025] Wherein, , and are the A, B and C three-phase currents measured by the three-phase current transformer;

[0026] X is the node feature vector: (3) ;

[0027] is the number of nodes; contains the time sequence samples of 1024 sampling points of the node ; the data is preprocessed by using the normalization method, and the formula is as follows:

[0028] (4) ;

[0029] Wherein, is the normalized data, is the minimum value in the original data, is the maximum value in the original data; after processing, the minimum value in the original data is mapped to 0, the maximum value is mapped to 1, and other values are linearly mapped between 0 and 1.

[0030] Optionally, S20 is specifically:

[0031] The processed adjacency matrix and node features are input into the built graph neural network model, spatial features and temporal features are extracted through GAT and LSTM respectively, different nodes are assigned different weights through the graph attention mechanism to aggregate more obvious spatial features, and the multiple gate structure units of the LSTM capture multiple dependency relationships of the time sequence signal through forgetting, updating and outputting. The time and space feature vectors are spliced together through a splicing layer, and the obtained prediction result and the real fault label are classified through a full connection layer and a sigmod classification layer. The loss value is calculated using the cross-entropy loss function, and the model parameters are updated through back propagation according to the size of the loss value, so that the model is optimized in the direction of smaller loss value; according to the probability Pi of each node fault output by the model, it is determined that there is a fault between which two nodes in the power distribution network, so as to determine the fault section; when the training result meets a certain accuracy, the model parameters are saved, and a trained model is obtained.

[0032] Optionally, S20 includes:

[0033] S201: Initialize the model: set the initial parameters of the graph attention network model, including weights and biases;

[0034] S202: Input data: input the adjacency matrix and node features of the constructed graph structure data as input data;

[0035] S203: Forward propagation: calculate the relevance between nodes through the graph attention mechanism and update the node features;

[0036] S204: Loss function: define a loss function to measure the difference between the model output and the actual fault location;

[0037] S205: Back propagation: calculate the gradient according to the loss function and update the model parameters;

[0038] S206: Iterative optimization: repeat the forward propagation and back propagation processes until the model loss reaches a preset threshold or the number of training reaches a set value;

[0039] S207: Model saving: save the trained model parameters for subsequent application.

[0040] Optionally, S203 includes:

[0041] The graph attention network structure is divided into two main parts, GAT and LSTM; among them, GAT extracts spatial features, and LSTM extracts time features; wherein GAT introduces an attention mechanism for the node and its adjacent node information aggregation to set the weight, thereby aggregating more significant features of each node; the input of each layer of GAT is a set of feature vectors of nodes ; N is the number of power grid topology nodes; x i is the feature vector of node i; F is the number of node features; its output is a new feature set: where x and have different dimensions;

[0042] First, the attention coefficient between each node is calculated, denoted as

[0043] (5);

[0044] where e ij is the attention coefficient of node j (the neighbor node of node i) to node i; is an attention function that maps the concatenated high-dimensional features to a real number; •||• is a feature concatenation operation; x i and x j are the input feature vectors of node i and node j; W is a shared parameter matrix of linear transformation; N i is the total number of first-order neighbor nodes of node i; the calculated attention coefficient is nonlinearly transformed and normalized, as follows

[0045] (6);

[0046] Then, the adjacent node vectors are weighted and summed according to the attention weight, to update the feature vector of the target node i, as follows

[0047] (7);

[0048] A multi-head attention mechanism is used to improve the performance of the model, with 4 attention heads set; that is, 4 independent attention mechanisms are used to calculate the attention weight and update the node features, and then the vectors updated by each attention mechanism are concatenated;

[0049] LSTM is a variant of recurrent neural network, which remedies the shortcomings of recurrent neural network that cannot remember information more than a long time ago and gradient explosion and disappearance; by stacking multiple gate structures to form an LSTM hidden layer, each gate structure contains an input gate, an output gate, and a forget gate, as well as two functions sigmod and tanh;

[0050] The calculation formula of each time step is as follows:

[0051] (8);

[0052] wherein is the output of the forget gate, , are the weights corresponding to the output of the previous gate structure and the input of the current period gate structure respectively, is the output of the previous period gate structure, is the input of the current period gate structure, is a bias term; after (sigmod) output of each dimension is between 0 and 1, 1 indicates that the original information is completely retained, and 0 indicates that the original information is completely removed;

[0053] (9);

[0054] (10);

[0055] In formula (9) (10) is the activation vector of the input gate, is the accumulated information at the current moment; the multiplication of the two obtains the state quantity to be updated;

[0056] (11);

[0057] In formula (11) is the state input quantity of the next gate structure, which is obtained by multiplying the previous gate structure state input quantity and the state information to be forgotten of the current structure and adding the state quantity to be updated of the current state ;

[0058] (12);

[0059] (13);

[0060] Formula (12) is the output gate activation vector of the LSTM, and formula (13) is the output of the LSTM, which multiplies the output gate activation vector and the state quantity of the current gate structure after tanh, to obtain the output.

[0061] Optionally, the S30 comprises:

[0062] S301: Fault probability calculation: using the trained model to predict the fault probability of the new graph structure data;

[0063] S302: Fault location: according to the output node failure probability Pi, determine the section where the fault is most likely to occur;

[0064] S303: Threshold setting: set a failure probability threshold, only when the failure probability of the node exceeds the threshold, it is considered that the node has a fault.

[0065] Optionally, the S204 comprises:

[0066] The graph attention network aggregates the input adjacency matrix and node features through GAT, and extracts the time sequence features of the zero sequence current of each node using LSTM in parallel with GAT; the feature vectors obtained by the two are spliced, so that each node has power distribution network topology spatial features and zero sequence current time sequence features, and the features are nonlinearly transformed through two fully connected layers, and finally the Sigmoid activation function is used to output the probability p of each node being a fault node i , so as to determine the fault section; the Sigmoid function formula is as follows:

[0067] (14) ;

[0068] wherein is the input value, and e is the base of natural logarithm;

[0069] The output prediction result is compared with the original label, the loss value is calculated using the Focal Loss loss function, the Focal Loss reduces the weight of easy-to-classify samples by increasing a regulating factor based on the balanced cross-entropy loss function, and focuses on the training of difficult samples, and its definition is as follows:

[0070] (15) ;

[0071] wherein, is the probability value predicted by the model, is a balance factor, is a regulating factor; by adjusting the values of and , the Focal Loss can pay more attention to difficult samples, thereby improving the performance of the model on unbalanced data sets.

[0072] Optionally, the S40 comprises:

[0073] S401: Data input: after preprocessing the field collected zero sequence current data, input it into the trained model;

[0074] S402: Fault prediction: the model outputs the failure probability of each node, and sorts according to the probability size;

[0075] S403: Fault line determination: according to the node with the highest fault probability, the fault line in the power distribution network is determined;

[0076] S404: Result output: output the fault line information.

[0077] The beneficial effects of the present application are: to provide a power distribution network fault line selection method considering the change of topological structure, the topological structure of the power distribution network is abstracted as a graph structure composed of nodes and edges, and the zero sequence current of the power distribution network is used as the feature vector of each node. The graph structure information is converted into an adjacency matrix, and is used as a model input together with the zero sequence current of each node, the spatial features of each node are aggregated using the graph attention mechanism of the GAT network, and the LSTM network extracts the time features of each node. The spatial features and time features are spliced to form new node features. The obtained node features are used to distinguish fault nodes and non-fault nodes, so as to achieve the effect of fault line positioning. Through the graph attention mechanism of GAT, different weights are assigned to different neighborhood nodes, and the fault line selection under the condition of the change of the topological structure of the power distribution network is realized.

[0078] Therefore, the power distribution network fault line selection method considering the change of topological structure provided by the present application can not only quickly and accurately locate the fault line, but also comprehensively consider the topological structure information of the power distribution network and adaptively adjust to cooperate with the dynamically changing structure of the power distribution network, thereby improving the efficiency and accuracy of fault line selection, and realizing effective fault monitoring in a wider power distribution network environment. BRIEF DESCRIPTION OF DRAWINGS

[0079] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0080] Figure 1 is the graph attention network model framework provided by the embodiment of the present application;

[0081] Figure 2 is the topological structure diagram of IEEE18 power distribution network provided by the embodiment of the present application;

[0082] Figure 3 is the flow chart of the power distribution network fault line selection method considering the change of topological structure provided by the embodiment of the present application. DETAILED DESCRIPTION

[0083] In order to make the objectives, characteristics and advantages of the present application more obvious and easy to understand, the technical solutions in the present embodiment will be clearly and completely described below in combination with the drawings in the present embodiment. Obviously, the following described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0084] The present embodiment provides a power distribution network fault line selection method considering topology changes, which is suitable for single-phase ground fault detection and line selection scene under the topology change of the power distribution network, and can improve the efficiency of power grid operation monitoring. The power distribution network fault line selection method considering topology changes is executed by a fault line selection device and realized by software and / or hardware.

[0085] Figure 3 The present embodiment provides a power distribution network fault line selection method considering topology changes, which is suitable for single-phase ground fault detection and line selection scene under the topology change of the power distribution network, and can improve the efficiency of power grid operation monitoring. The power distribution network fault line selection method considering topology changes is executed by a fault line selection device and realized by software and / or hardware.

[0086] Referring to Figure 3 , the power distribution network fault line selection method considering topology changes comprises the following steps:

[0087] S10: Data preprocessing: by collecting zero sequence current data and converting it into graph structure data, input information representing the topology and current characteristics of the power distribution network is provided for the model;

[0088] S101: Data acquisition: zero sequence current data is acquired from the power distribution network site;

[0089] S102: Data cleaning: remove noise and outliers in the collected zero sequence current data;

[0090] S103: Feature extraction: extract key features such as current amplitude, phase angle, etc. from the cleaned data;

[0091] S104: Construct graph structure: convert the node and branch information of the power distribution network into graph structure data, where the nodes represent the various parts of the power distribution network and the edges represent the connection relationship between the nodes;

[0092] First, the topology connection relationship of the measurement nodes of the power distribution network is obtained, and a node adjacency matrix is made according to formula (1). By measuring the data of the zero sequence current transformer in the field, the required zero sequence current data samples for the experiment are obtained.

[0093] From the perspective of graph, without considering the internal components and line characteristics, the node network of the power distribution network is abstracted into graph structure data G=(X,A) of nodes and edges; the adjacency matrix A describes the connection relationship of each node of the power distribution network, wherein the row and column numbers of the adjacency matrix correspond to the nodes, as shown in formula (1):

[0094] (1);

[0095] The branch zero sequence current is used as the feature vector of the branch head node. In a power system, the zero sequence current refers to the current generated when the vector sum of three-phase currents is not zero in a three-phase system. The zero sequence current is usually related to ground faults. In actual scenarios, for cable lines, a zero sequence current transformer is a device specially used to detect zero sequence currents. It is usually installed on the neutral point of a three-phase power system and can detect zero sequence currents generated due to faults or unbalanced loads. A zero sequence current transformer is generally used to collect zero sequence currents. The current measurement of overhead lines usually uses three-phase current transformers, which are installed on each line of the three-phase system. By measuring the three-phase currents, the zero sequence current can be calculated. The calculation method is as follows:

[0096] (2);

[0097] wherein, , and are the A, B, and C three-phase currents measured by the three-phase current transformers.

[0098] X is the node feature vector: (3);

[0099] is the number of nodes; contains the time sequence samples of the 1024 sampling points of the node . A normalization method is used to preprocess the data, and the formula is as follows:

[0100] (4);

[0101] wherein, is the normalized data, is the minimum value in the original data, is the maximum value in the original data. After this processing, the minimum value in the original data is mapped to 0, the maximum value is mapped to 1, and other values are linearly mapped between 0 and 1. This helps to reduce the influence of outliers on the model.

[0102] Optionally, the sample data obtained by field measurement is used to normalize the fault current signal of each measurement point using formula (4) to obtain processed experimental data, which is used as the node feature data of each node. The nodes on both sides of the fault section of the IEEE18 node network (as shown in Figure 2 ) are taken as fault nodes, and the labels of the fault nodes are set to 1, and the labels of the non-fault nodes are set to 0.

[0103] S20: Model training: using graph attention network to learn the adjacency matrix and node features until the model can accurately predict the probability of single-phase grounding fault of the power distribution network;

[0104] S20 is specifically: input the processed adjacency matrix and node features into the built graph neural network model, extract spatial features and time features through GAT and LSTM respectively, assign different weights to different nodes through the graph attention mechanism to aggregate more obvious spatial features, and the multiple gate structure units of LSTM capture multiple dependency relationships of time series signals through forgetting, updating and outputting, concatenate the time and space feature vectors through a concatenation layer, and through a fully connected layer and a sigmod classification layer, calculate the loss value using the cross-entropy loss function between the predicted result and the real fault label, update the model parameters according to the size of the loss value, and optimize the model to a smaller loss value; according to the probability P i of each node fault output by the model, determine whether there is a fault between two nodes in the power distribution network, and thus determine the fault section; save the model parameters when the training result meets a certain accuracy to obtain the trained model;

[0105] Specifically:

[0106] S201: Initialize the model: set the initial parameters of the graph attention network model, including weights and biases;

[0107] The zero sequence current signal features of each node of the power distribution network are extracted and fused with the topology structure of the power distribution network using graph theory knowledge, and a power distribution network grounding fault line selection model based on a graph attention network is established. The input of the graph attention network includes two parts: one is the adjacency matrix representing the topology structure of the power distribution network, and the other is the node features corresponding to each node, i.e. the zero sequence current.

[0108] S202: Input data: input the adjacency matrix and node features of the constructed graph structure data as input data;

[0109] S203: Forward propagation: calculate the correlation between nodes through the graph attention mechanism and update the node features;

[0110] Referring to Figure 1 , the graph attention network structure is divided into two main parts, GAT and LSTM. Among them, GAT extracts spatial features, and LSTM extracts time features. GAT introduces an attention mechanism on the basis of GCN to set weights for the information aggregation of nodes and their adjacent nodes, thereby aggregating more significant features of each node. The input of each layer of GAT is a set of node feature vectors . N is the number of nodes in the power distribution network topology; x iis the feature vector of node i; F is the number of features of the nodes. Its output is a new feature set: where x and have different dimensions.

[0111] First, the attention coefficients between nodes are calculated, denoted as

[0112] (5).

[0113] where e ij is the attention coefficient of node j (a neighbor node of node i) to node i; is the attention function, which maps the concatenated high-dimensional features to a real number; •||• is the feature concatenation operation; x i and x j are the input feature vectors of node i and node j; W is the shared parameter matrix of linear transformation; N i is the total number of first-order neighbor nodes of node i. The calculated attention coefficients are nonlinearly transformed and normalized, as follows

[0114] (6).

[0115] Then, the neighboring node vectors are weighted and summed according to the attention weights, to update the feature vector of the target node i, as follows

[0116] (7).

[0117] A multi-head attention mechanism is used to improve the performance of the model, with the number of attention heads set to 4. That is, through 4 independent attention mechanisms, the attention weights are calculated and the node features are updated, and then the vectors updated by each attention mechanism are concatenated.

[0118] When the topology of the power distribution network changes, for example, the addition or deletion of nodes or branches, this will cause the network's adjacency matrix and node feature vectors to change accordingly. In this case, the Graph Attention Network (GAT) can automatically adjust the attention weights between nodes through its attention mechanism, as shown in equation (4), to adapt to these changes. Since the operation of GAT is independent for each vertex, its core parameters W and mapping function a mainly depend on the features of the nodes, which means that GAT can flexibly handle changes in the topology structure, maintaining the applicability and effectiveness of the model. In short, the design of GAT allows it to dynamically adapt to changes in the network structure without the need for retraining or complex adjustments.

[0119] LSTM is a variant of recurrent neural network, which makes up the shortcomings of recurrent neural network that cannot remember information more than a long time ago and the gradient will explode or disappear. By adopting a variety of gate structure to stack the LSTM hidden layer, each gate structure contains an input gate, an output gate and a forget gate, and there are two functions sigmod and tanh.

[0120] The calculation formula of each time step is as follows:

[0121] (8) ;

[0122] In the formula is the output of the forget gate, ,are the weights corresponding to the output of the previous gate structure and the input of the current period gate structure respectively, is the output of the previous period gate structure, is the input of the current period gate structure, is the bias term; after (sigmod) output, the value of each dimension is between 0 and 1, 1 means completely retaining the original information, and 0 means completely removing the original information.

[0123] (9) ;

[0124] (10) ;

[0125] In the formula (9) (10) is the activation vector of the input gate, is the accumulated information at the current moment. Multiply the two to get the state quantity to be updated.

[0126] (11) ;

[0127] In the formula (11) is the state input quantity of the next gate structure, which is obtained by multiplying the state input quantity of the previous gate structure and the state information that needs to be forgotten in the current structure and adding the state quantity that needs to be updated in the current state .

[0128] (12) ;

[0129] (13) ;

[0130] Formula (12) is the output gate activation vector of LSTM, and formula (13) is the output of LSTM, which is obtained by multiplying the activation vector of the output gate and the state quantity of the current gate structure after tanh.

[0131] S204: Loss function: define a loss function to measure the difference between the model output and the actual fault location;

[0132] The graph attention network aggregates the input adjacency matrix and node features through GAT, and at the same time uses LSTM parallel to GAT to extract the time sequence features of the zero sequence current of each node. The feature vectors obtained by the two are spliced, so that each node has both the spatial features of the power distribution network topology and the time sequence features of the zero sequence current, and two fully connected layers are used to perform nonlinear transformation on the features, and finally the Sigmoid activation function is used to output the probability p of each node being a fault node i , so as to determine the fault section. The formula of the Sigmoid function is as follows:

[0133] (14) ;

[0134] wherein is the input value, and e is the base of natural logarithm.

[0135] The output prediction result is compared with the original label, and the loss value is calculated using the Focal Loss loss function. Focal Loss increases a regulation factor to reduce the weight of easy-to-classify samples on the basis of balancing the cross-entropy loss function, and focuses on the training of difficult samples, and its definition is as follows:

[0136] (15) ;

[0137] wherein, is the probability value predicted by the model, is the balance factor, is the regulation factor. By adjusting the values of and , Focal Loss can pay more attention to difficult samples, thereby improving the performance of the model on unbalanced data sets.

[0138] S205: Back propagation: calculate the gradient according to the loss function, and update the model parameters;

[0139] Back propagation updates the gradient, so that the performance of the model approaches the direction of smaller loss value.

[0140] S206: Iterative optimization: repeat the forward propagation and back propagation process until the model loss reaches the preset threshold or the number of training reaches the set value.

[0141] S207: Model saving: Save the trained model parameters for subsequent application.

[0142] S30: Fault judgment: Based on the node fault probability output by the model, set a threshold to identify and locate the specific fault line;

[0143] S301: Fault probability calculation: Use the trained model to predict the fault probability of new graph structure data;

[0144] S302: Fault location: According to the output node fault probability Pi, determine the section where the fault is most likely to occur;

[0145] S303: Threshold setting: Set a fault probability threshold, only when the node's fault probability exceeds the threshold, it is considered that the node has a fault.

[0146] S40: Model application: Input the field zero sequence current data into the trained model to quickly and accurately judge the fault location of the distribution network;

[0147] S401: Data input: After preprocessing the field collected zero sequence current data, input it into the trained model;

[0148] S402: Fault prediction: The model outputs the fault probability of each node, which is sorted according to the probability size;

[0149] S403: Fault line determination: According to the node with the highest fault probability, determine the fault line in the distribution network;

[0150] S404: Result output: Output the fault line information for reference and processing by operation and maintenance personnel.

[0151] The power distribution network fault line selection method considering topology structure changes provided by the embodiment has the following beneficial effects:

[0152] ① Improve fault detection accuracy: By combining graph attention network (GAT) and long short-term memory network (LSTM), this method can capture both spatial and temporal features of the distribution network, thus more accurately identifying and locating single-phase ground faults.

[0153] ② Adapt to topology structure changes: This method can adapt to changes in the topology structure of the distribution network, such as the addition or subtraction of nodes or branches, without the need to retrain the model or make complex adjustments, improving the flexibility and adaptability of the model.

[0154] ③ Reduce operation and maintenance costs: By quickly and accurately determining the fault location, the workload of operation and maintenance personnel can be reduced, the operation and maintenance costs can be reduced, and the efficiency of power grid operation monitoring can be improved.

[0155] (4) Enhance the reliability of the power grid: timely and accurate detection and positioning of faults can quickly take measures to restore normal operation of the power grid, reduce power outage time, and improve the reliability of the power grid and user satisfaction.

[0156] (5) Handle imbalanced datasets: By using the Focal Loss loss function, the method can better handle imbalanced datasets, focusing on difficult sample training, and improving the performance of the model on imbalanced datasets.

[0157] The above examples are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent substitutions for part of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fault line selection method for a power distribution network considering topology changes, characterized in that, Comprise: S10: data preprocessing: by collecting zero sequence current data and converting into graph structure data, providing input information for the model to represent the topology and current characteristics of the power distribution network; S20: model training: using graph attention network to learn the adjacency matrix and node features until the model can accurately predict the probability of single-phase ground fault of the power distribution network; S30: fault judgment: based on the node fault probability output by the model, set a threshold to identify and locate the specific fault line; S40: model application: input the field zero sequence current data into the trained model to quickly and accurately judge the fault location of the power distribution network; Among them, S10 comprises: S101: data collection: collect zero sequence current data from the power distribution network site; S102: data cleaning: remove noise and outliers from the collected zero sequence current data; S103: feature extraction: extract key features from the cleaned data; S104: construct graph structure: convert the node and branch information of the power distribution network into graph structure data, where the node represents each part of the power distribution network, and the edge represents the connection relationship between the nodes; S104 comprises: First, obtain the topological connection relationship of the measurement nodes of the power distribution network, and make a node adjacency matrix according to formula (1); obtain the zero sequence current data samples needed for the experiment by measuring the data of the zero sequence current transformer on site; The sample data obtained by on-site measurement is normalized for each measurement point using formula (4) to obtain the processed experimental data, which is used as the node feature data of each node; the nodes on both sides of the fault section of the IEEE18 node network are regarded as fault nodes, and the labels of the fault nodes are set to 1, and the labels of the non-fault nodes are set to 0: The adjacency matrix A describes the connection relationship of each node of the power distribution network, wherein the row and column numbers of the adjacency matrix correspond to the nodes, as shown in formula (1): （1）； The branch zero sequence current is used as the feature vector of the branch head node; The data is preprocessed by using the normalization method, and the formula is as follows: （4）； wherein I is the original current data, is the normalized data, is the minimum value in the original data, is the maximum value in the original data; S20 specifically comprises: The processed adjacency matrix and node features are input into the built graph neural network model, the spatial features and temporal features are extracted through GAT and LSTM respectively, the different nodes are allocated different weights through the graph attention mechanism to aggregate the more obvious spatial features, and the multiple gate structure units of the LSTM capture the multiple dependency relationships of the time sequence signals through forgetting, updating and outputting. Through a splicing layer, the time and space feature vectors are spliced together, and then through a full connection layer and a sigmod classification layer, the loss value is calculated by using the cross-entropy loss function for the prediction result and the real fault label. According to the size of the loss value, the model parameters are updated through back propagation to optimize the model in the direction of smaller loss value; according to the probability Pi of each node fault output by the model, it is determined that there is a fault between which two nodes of the power distribution network, so as to determine the fault section; when the training result meets a certain accuracy, the model parameters are saved to obtain the trained model.

2. The power distribution network fault sectioning method of claim 1, wherein, The key features extracted from the data include current amplitude and / or phase angle.

3. The power distribution network fault sectioning method of claim 2, wherein, S20 comprises: S201: Initialize the model: set the initial parameters of the graph attention network model, including weights and biases; S202: Input data: input the adjacency matrix and node features of the constructed graph structure data as input data; S203: Forward propagation: calculate the relevance between nodes through the graph attention mechanism and update the node features; S204: Loss function: define a loss function to measure the difference between the model output and the actual fault location; S205: Backpropagation: calculate the gradient according to the loss function and update the model parameters; S206: Iterative optimization: repeat the forward propagation and backpropagation process until the model loss reaches the preset threshold or the number of training reaches the set value; S207: Model saving: save the trained model parameters for subsequent application.

4. The power distribution network fault sectioning method of claim 3, wherein, The S30 includes: S301: Fault probability calculation: use the trained model to predict the fault probability of new graph structure data; S302: Fault location: determine the section where the fault is most likely to occur according to the output node fault probability Pi; S303: Threshold setting: set a fault probability threshold, only when the node's fault probability exceeds the threshold, it is considered that the node has a fault.

5. The power distribution network fault sectioning method of claim 4, wherein, The S204 includes: The graph attention network aggregates the input adjacency matrix and node features through GAT to obtain spatial features, and extracts the time sequence features of the zero sequence current of each node through LSTM in parallel with GAT; the feature vectors obtained by the two are spliced, so that each node has the spatial features of the power distribution network topology and the time sequence features of the zero sequence current at the same time, the features are nonlinearly transformed through two fully connected layers, and finally the probability p that each node is a fault node is output through a Sigmoid activation function i , so as to judge the fault section; the formula of the Sigmoid function is as follows: （14）； wherein is the input value, e is the base of the natural logarithm; Compare the output prediction results with the original labels, and calculate the loss value using the Focal Loss loss function. Focal Loss increases a regulation factor to reduce the weight of easy-to-classify samples based on the balanced cross-entropy loss function, focusing on the training of difficult samples, and its definition is as follows: （15）； wherein, is the model predicted probability value, is the balancing factor, is the adjustment factor; by adjusting and the values of, focal loss can pay more attention to difficult samples, thus improving the performance of the model on imbalanced datasets.

6. The power distribution network fault sectioning method of claim 5, wherein, The S40 includes: S401: Data input: input the preprocessed zero sequence current data collected on site into the trained model; S402: Fault prediction: the model outputs the fault probability of each node, and sorts according to the probability size; S403: Fault line determination: determine the fault line in the distribution network according to the node with the highest fault probability; S404: Result output: output the fault line information.

Images