Distribution network fault section positioning and interpretability analysis method fusing space-time diagram information

Abstract

The invention discloses a distribution network fault section positioning and interpretability analysis method fusing space-time diagram information, and belongs to the technical field of distribution network fault diagnosis and artificial intelligence deep learning. Comprising the following steps: describing a mapping relation between spatial-temporal characteristics and fault sections by constructing a spatial-temporal diagram information fault representation structure fusing time continuity and spatial integrity; fault features are extracted from two dimensions of time feature capture and space feature perception, and a power distribution network fault positioning model fused with space-time diagram information is established; and designing an interpretability analysis and verification module, carrying out post interpretability analysis on a model decision basis and an internal working mechanism thereof, and realizing verification of a fault positioning result. According to the method, a high-precision fault positioning task in complex operation scenes such as new energy output fluctuation and topological change can be realized, the model and a fault prediction principle are planed and described, and technical support is provided for application of the fault positioning method based on deep learning in an actual system.

Description

Technical Field

[0001] The present invention relates to the field of distribution network fault diagnosis and artificial intelligence deep learning technology, and in particular to a distribution network fault section positioning and explainability analysis method integrating spatiotemporal graph information. Background Art

[0002] In complex operating scenarios such as fluctuations in renewable energy output and frequent topology reconstruction, the safe and reliable operation of the distribution system faces new challenges. Therefore, quickly and accurately locating faults can provide effective guidance for operation and maintenance personnel, greatly shorten the fault handling time, and is of great significance to improving the reliability of system operation.

[0003] Traditional distribution network fault location methods mainly include impedance method, traveling wave method and artificial intelligence method. The impedance method aims to calculate the distance from the measurement point to the fault point by the ratio of loop impedance to line unit impedance, and is one of the easiest methods to implement in field operations. However, when applying the impedance method, the network topology must be equivalent first. Once the topology changes, the equivalent calculation must be performed again, which is difficult to adapt to when the distribution network topology changes frequently. The traveling wave method uses the propagation time of the fault transient traveling wave signal in the line and the traveling wave signal to achieve fault location, which is basically not affected by the transition resistance and neutral point grounding method. However, the fault traveling wave is easily affected by the complex topology of the distribution network, and signal attenuation or distortion occurs during the transmission and refraction and reflection process, resulting in reduced fault location accuracy.

[0004] At present, the artificial intelligence solutions used for distribution network fault location mainly include artificial neural networks, convolutional neural networks, long short-term memory neural networks and other composite models. Although the improved solutions of the fault location model based on deep learning have achieved good application results, there are still certain limitations. That is, the above-mentioned solutions usually directly splice multiple time series signals from different measurement units into a longer two-dimensional signal, or construct them into a two-dimensional image and send them to the deep learning network without considering spatial information, which leads to the spatial characteristics of the interaction between devices being ignored, resulting in poor generalization performance of existing solutions in scenarios where the network topology changes frequently. At the same time, this type of method fails to conduct an explainable analysis of the "black box problem" that is prevalent in artificial intelligence models, and the lack of explainability has become one of the main obstacles limiting deep learning in safety-sensitive tasks such as distribution network fault diagnosis. Summary of the invention

[0005] The purpose of the present invention is to provide a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information, which solves the problems of certain limitations and lack of interpretability.

[0006] To achieve the above object, the present invention provides a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information, comprising the following steps:

[0007] S1. Construct a fault representation structure of spatiotemporal graph information, integrate fault feature information from the perspective of temporal continuity and spatial integrity, and enhance the recognition ability of the classification model;

[0008] S2. Establish a distribution network fault location model integrating spatiotemporal graph information;

[0009] S3. Design an explainable analysis verification module to conduct explainable analysis on the decision-making basis and internal working mechanism.

[0010] Preferably, in step S1, the time continuity is to mine the electrical variation of the fault information based on the continuous time series of the voltage and current collected at the measuring point;

[0011] Spatial integrity refers to the on-off change of the link relationship of topological nodes during the operation of the distribution network.

[0012] Preferably, in step S1, the constructed spatiotemporal graph information fault characterization structure expression is:

[0013]

[0014] Among them, p j is the measuring point numbered j in the distribution line; They represent the three-phase voltage and three-phase current collected at the measuring point at p j The instantaneous value of a moment.

[0015] Preferably, in step S2, the fault location model includes a time series information capture module and a spatial feature perception module;

[0016] In order to capture the fault characteristics contained in the time series, the timing information capture module uses the gated convolution unit as the main structure of the timing information capture module, and realizes feature enhancement by merging the features on continuous time segments. The model structure consists of a gated convolution unit composed of two traditional one-dimensional convolution layers, and its mathematical expression is as follows:

[0017]

[0018] Where: H l and H l+1 are the input and output of the Gated-CNN module at layer l, respectively; and is the convolution kernel of the same dimension; and is the bias of the model; is the convolution operation; σ1(·) is the gating function, which takes the Sigmoid activation function;

[0019] In order to better represent the spatial dependency and propagation mode of fault points in the distribution network, the spatial feature perception module uses the distribution network topology structure based on the graph convolutional neural network to capture the fault impact by aggregating node features. Its mathematical expression is:

[0020]

[0021] Where: X l ∈R N×C is the input matrix of the lth hidden layer, N is the number of measurement points, C is the number of sampling points of each measurement point; σ2(·) is the ReLU activation function; represents the sum of the adjacency matrix A and the identity matrix; Represents the degree matrix corresponding to the adjacency matrix A; W∈R C×Q , represents the weight matrix of the convolutional layer, Q is the number of hidden layer nodes. Finally, the lth hidden layer output f(A,X l ).

[0022] Preferably, usually, the distribution network reconstruction (such as changes in node connectivity caused by switch operations) occurs in local areas, and GCN, due to its natural property of aggregating local neighborhood information, can adjust the information transmission path in a timely manner without large-scale adjustments to the entire model, and thus has a strong adaptability to new topological structures.

[0023] Preferably, in step S3, the established interpretable analysis verification module combines graph data with gradient information to identify the feature nodes that the model focuses on in the decision-making process, thereby evaluating the importance of graph nodes. The specific steps are as follows:

[0024] S31, class activation loss update

[0025] Assuming that the number of measurement points in the topology is n, the number of sampling channels of each measurement point is m, the sampling length is l, and the fault location task of c sections of the distribution network needs to be completed, the fault representation structure of the spatiotemporal graph information is X = {X1, X2, …, X n}, X∈R n×m×l As the input of the interpretability stage, its label is Y = {Y1,Y2,…,Y n}. Among them, X i represents the timing signal collected at the i-th measurement point, Y i Indicates the fault section corresponding to the signal.

[0026] Construct a new STGCN network with the same structure as the trained STGCN network, retain the training weight parameters, and define a new class activation loss function L CAM As shown in formula (4):

[0027]

[0028] Among them, Z i ={Z1,Z2,Z3,…,Z n}, is X i The final classification output after inputting the model; the maximum value of the final classification output obtained by the argmax(·) function. The class activation loss function L defined CAM As the loss function of STGCN, the model is trained so that the prediction result Z of each measurement point i As close to Y as possible i .

[0029] S32. Spatial weight map generation

[0030] The fault representation structure X of the spatiotemporal graph information i Input L CAM The new STGCN network with loss function calculates the weight contribution of the sth measurement point to the final classification result As shown in formula (5):

[0031]

[0032] Among them, d l+1 is the output dimension of the graph convolutional layer; y e Refers to the probability that the classification output is the e-th segment;

[0033] The physical meaning of is the importance of the i-th feature point of the s-th measurement point to the classification result. When the e-th section fails, the weight vector of each measurement point can be obtained. As shown in formula (6):

[0034]

[0035] According to the weight vector of each measurement point The topological adjacency matrix A is used to determine the impact of node measurement information on the positioning results during the model classification process through the size of the spatial weight of each node. If topological reconstruction occurs, the connection relationship of the spatial weight graph changes. The STGCN module can be retrained and the confusion matrix of the classification results can be analyzed to verify the importance of the node to the specific section fault.

[0036] Therefore, the present invention adopts a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information of the above structure, which has the following beneficial effects:

[0037] The present invention has good generalization ability under complex operating scenarios such as renewable energy output fluctuations, noise interference and topology reconstruction, and can provide an effective solution to the problem of fault location in dynamic distribution networks. At the same time, the interpretable analysis module designed by the present invention can quantify the contribution of the information of each measurement point in the output result, and verify the rationality of the fault section location by reconstructing the spatial weight map, providing a new research idea for the interpretability research in the field of distribution network fault diagnosis.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The topology diagram of the 10kV distribution network model for simulation verification;

[0040] Figure 2 This is a flow chart of the fault section location solution;

[0041] Figure 3 This is a diagram of the model training process;

[0042] Figure 4 is the spatial weight map of line L4B2 when it fails;

[0043] Figure 5 The spatial weight map of line L4B2 when it fails after the adjacency matrix is ​​modified. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0045] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0046] Example

[0047] The present invention provides a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information, comprising the following steps:

[0048] S1, constructing a fault representation structure of spatiotemporal graph information;

[0049] The constructed spatiotemporal graph information fault representation structure integrates fault feature information from the perspective of temporal continuity and spatial integrity, and enhances the recognition ability of the classification model. The details are as follows:

[0050] Time continuity mainly refers to the electrical changes in fault information mined by the continuous time series of voltage and current collected at the measurement points.

[0051] Spatial integrity mainly refers to the switching changes in the topological node link relationship during the operation of the distribution network.

[0052] The constructed spatiotemporal graph information fault representation structure can be expressed as formula (1):

[0053]

[0054] Among them, p j is the measuring point numbered j in the distribution line; They represent the three-phase voltage and three-phase current collected at the measuring point at p j The instantaneous value of a moment.

[0055] S2. Establish a distribution network fault location model integrating spatiotemporal graph information;

[0056] The established fault location model mainly includes the time series information capture module and the spatial feature perception module. The specific modules are described as follows:

[0057] Timing information capture module:

[0058] In order to capture the fault features contained in the time series, the gated convolution unit is selected as the main structure of the time series information capture module, and feature enhancement is achieved by merging the features on continuous time segments. The model structure consists of a gated convolution unit composed of two traditional one-dimensional convolution layers, and its mathematical expression is as shown in formula (2):

[0059]

[0060] Where: H l and H l+1 are the input and output of the Gated-CNN module at layer l, respectively; and is the convolution kernel of the same dimension; and is the bias of the model; is the convolution operation; σ1(·) is the gating function, which takes the Sigmoid activation function.

[0061] Spatial feature perception module

[0062] In order to better represent the spatial dependency and propagation pattern of fault points in the distribution network, the topological structure of the distribution network is used to capture the fault impact by aggregating node features based on the graph convolutional neural network. The mathematical expression of the model is as follows (3):

[0063]

[0064] Where: X l ∈R N×C is the input matrix of the lth hidden layer, N is the number of measurement points, C is the number of sampling points of each measurement point; σ2(·) is the ReLU activation function; It represents the sum of the adjacency matrix A and the identity matrix, that is, the node can obtain the comprehensive information of itself and the surrounding nodes; Represents the degree matrix corresponding to the adjacency matrix A; W∈R C×Q , represents the weight matrix of the convolutional layer, Q is the number of hidden layer nodes. Finally, the lth hidden layer output f(A,X l ).

[0065] Typically, distribution network reconstruction (such as changes in node connectivity caused by switch operations) occurs in local areas. Due to its natural property of aggregating local neighborhood information, GCN can adjust the information transmission path in a timely manner without large-scale adjustments to the entire model, and thus has a strong adaptability to new topological structures.

[0066] S3. Design an explainable analysis verification module to conduct explainable analysis on the decision-making basis and internal working mechanism.

[0067] The interpretable analysis verification module combines graph data with gradient information to identify the feature nodes that the model focuses on during the decision-making process, thereby evaluating the importance of graph nodes. The specific steps are as follows:

[0068] S31, class activation loss update

[0069] Assuming that the number of measurement points in the topology is n, the number of sampling channels of each measurement point is m, the sampling length is l, and the fault location task of c sections of the distribution network needs to be completed, the fault representation structure of the spatiotemporal graph information is X = {X1, X2, …, X n}, X∈R n×m×l As the input of the interpretability stage, its label is Y = {Y1,Y2,…,Y n}. Among them, X i represents the timing signal collected at the i-th measurement point, Y i Indicates the fault section corresponding to the signal.

[0070] Furthermore, a new STGCN network with the same structure as the trained STGCN network and the same training weight parameters is constructed, and a new class activation loss function L is defined. CAM As shown in formula (4):

[0071]

[0072] Among them, Z i ={Z1,Z2,Z3,…,Z n}, is X i The final classification output after inputting the model; the maximum value of the final classification output obtained by the argmax(·) function. The class activation loss function L defined CAM As the loss function of STGCN, the model is trained so that the prediction result Z of each measurement point i As close to Y as possible i .

[0073] S32. Spatial weight map generation

[0074] The fault representation structure X of the spatiotemporal graph information i Input L CAM The new STGCN network with loss function calculates the weight contribution of the sth measurement point to the final classification result As shown in formula (5):

[0075]

[0076] Among them, d l+1 is the output dimension of the graph convolutional layer; y e Refers to the probability that the classification output is the e-th segment; The physical meaning of is the importance of the i-th feature point of the s-th measurement point to the classification result. When the e-th section fails, the weight vector of each measurement point can be obtained. As shown in formula (6):

[0077]

[0078] According to the weight vector of each measurement point The topological adjacency matrix A is used to determine the impact of node measurement information on the positioning results during the model classification process through the size of the spatial weight of each node. If topological reconstruction occurs, the connection relationship of the spatial weight graph changes. The STGCN module can be retrained and the confusion matrix of the classification results can be analyzed to verify the importance of the node to the specific section fault.

[0079] Build in MATLAB / Simulink Figure 1The 10kV distribution network simulation model shown in the figure includes power supply, transformer, line, load and distributed power supply. Among them, the main transformer adopts Dyn11 connection method, the transformation ratio is 110kV / 10.5kV, the neutral point adopts arc suppression coil grounding method, and the compensation degree is 8%; the length and type of lines L1B1, L1B2, ..., L6B3 are marked in the figure, and the line parameters are shown in Table 1; the branch connection is changed by the opening and closing state of the connecting switch S1~S6 to realize the topology reconstruction. Under normal circumstances, only switches S1~S3 are closed; distributed power sources DG1~DG3 are connected to bus 1~3 respectively. It is a μPMU device with a sampling frequency of 10kHz, which can collect high-precision voltage / current waveform data with time alignment. The data within the 60ms time window before and after the start judgment action is uniformly intercepted, including 1 cycle before the fault and 2 cycles after the fault.

[0080] Table 1 Line parameters

[0081]

[0082] According to the above simulation settings, different fault types, transition resistances, initial fault phase angles, load levels, fault lines and relative positions are set to simulate fault scenarios. The simulation sample parameters of specific fault cases are shown in Table 2, and a total of 1620 fault samples are generated. In addition to the above fault samples, normal operating samples are obtained by changing different load levels. Starting from the moment when the system is completely stable under each normal operating condition, with a time window length of 60ms, the window is slid back 100 times every 10 measurement points. The specific normal operating sample parameters are shown in Table 3, and a total of 300 normal operating samples are generated. After the order of the above 1920 samples is shuffled, they are randomly divided into training sets and test sets at a ratio of 4:1, and the F1 score is used to evaluate the classification effect of the fault location model. The calculation formula is:

[0083]

[0084] Among them: True Positive (TP) refers to samples that are actually positive and predicted to be positive; False Positive (FP) refers to samples that are actually negative and predicted to be positive; False Negative (FN) refers to samples that are actually positive and predicted to be negative. The F1 score can take into account the precision (Precision, P) and recall (Recall, R) of the classification model, and is used to comprehensively evaluate the positioning accuracy of the model. It can be considered that the larger the F1 value, the better the model classification performance.

[0085] Table 2 Line parameters

[0086]

[0087]

[0088] The present invention is specifically implemented by the following technical scheme: a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information, comprising the following steps:

[0089] 1. Raw data acquisition and preprocessing

[0090] For the above voltage or current time series Y = {y1, y2, ..., y k}(y k The data is normalized and scaled to the interval [-1, 1], as shown in formula (8):

[0091]

[0092] Where: y max is the maximum value in the time series Y; min is the minimum value in the sequence; is the value after data normalization, the original time series is transformed into

[0093] At the same time, for Figure 1 For the distribution network shown in the figure, the adjacency matrix A1 can be constructed based on the spatial relationship of the five measurement points to describe the connection relationship between the points in the topology diagram. When the switch S1 changes from closed to open and S2 changes from open to closed, the adjacency matrix changes to A2, as shown in formula (9):

[0094]

[0095] 2. Model training and performance improvement

[0096] After preprocessing the raw data and generating the adjacency matrix, the processed data set is used to train a deep learning model with better performance offline. In the training stage, the data set is divided into a training set and a test set according to a certain ratio. Among them, the training set is used to train the network parameters of the model, and the hyperparameters are adjusted according to the classification effect of the model to select the optimal network architecture and model performance. During the training process, cross entropy is used as the loss function to quantify the difference between the model output results and the true label; the Adam optimizer is used to calculate the minimum loss function, optimize the weights and biases of each layer of the model, and test the model convergence. This paper takes the initial learning rate as 10 -5 , the attenuation coefficient is 0.5, and the attenuation period is 3. Finally, the fault location model that has been trained and has good performance is saved for subsequent applications.

[0097] 3. Online Application and Explainable Analysis

[0098] During online operation, each measuring point collects the three-phase voltage and current signals of each line. At the same time, in order to avoid repeated detection and waste of computing resources, this paper uses phase current mutation and zero-sequence overcurrent as the starting criteria, as shown in equations (10) and (11).

[0099]

[0100] In formula (8), is the phase current channel at any time t k The effective value of N is the interval period; I n1 is the maximum value of each phase current mutation during normal operation. Formula (9) is set to prevent the lack of sensitivity of the phase current mutation starting element during high-resistance grounding fault. They are the three-phase currents at any time t k The instantaneous value of n2 is the reference current, here we take 1A; k set1 and k set2 are the starting thresholds, which are taken as 0.2 and 0.1 respectively in this paper.

[0101] If the current signal of a certain measurement point meets the requirements of the above-mentioned start-up criteria, the voltage and current collected at each measurement point are collected in the form of formula (1) and data preprocessed based on the start-up time of the criteria. The fault location model that has been saved after offline training is called for judgment to obtain the fault diagnosis result. In the interpretable analysis stage, the classification mechanism of the fault location model is analyzed using the interpretable analysis module. By calculating the input spatiotemporal feature attribution value, the influence of the input information of a single measurement point on the output result is obtained; combined with the spatial weight map, the correlation between the measurement information and the output category is analyzed from a global perspective, and the decision-making basis of the model under different spatial feature distributions is intuitively explained, so as to understand the internal working mechanism of the model.

[0102] The proposed distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information is as follows: Figure 2 shown.

[0103] pass Figure 3 The performance of the model on the training set and the test set during the training process is shown. It can be seen that at the 160th iteration, the model began to converge and stabilize, and the accuracy on the test set can reach 92.9%, indicating that the proposed scheme has a very strong fault location capability and is not affected by the fault type, transition resistance, and fault initial phase angle.

[0104] Taking the line L4B2 fault as an example, we select the data under a certain operating condition and input it into the interpretable analysis module to obtain the following: Figure 4The spatial weight diagram is shown in . It can be seen that the measurement points P4, P11 and P12 have larger weights, that is, the data features of these three measurement points have made a greater contribution to the judgment of the line fault, and the data contribution at P11 is the largest. Analysis of the above results shows that the positions of the three measurement points are close to the corresponding fault line, and the fault propagation path is shorter than that of other measurement points, so the most obvious fault characteristics can be obtained, which makes a greater contribution to the model's decision-making on the corresponding position fault. When other lines have faults under certain operating conditions, the weight values ​​of the measurement points are shown in Table 3.

[0105] Table 3 Weight values ​​of measurement points when each line fails

[0106]

[0107]

[0108] When topology reconstruction occurs (i.e., S2 is opened and S5 is closed), the measuring point P is cut off. 11 With P 12 The connection relationship between them is modified, the input adjacency matrix is ​​modified, and the spatiotemporal graph structure data is input into the STGCN module for testing. Except for the case where 3 additional line L4B2 fault samples (the real label is 9) are misjudged as line L4B3 faults (the predicted result is 10), the judgment results of the other lines remain unchanged.

[0109] Further analysis of the model gives the internal principle of the above decision, and derives the spatial weight map corresponding to the samples with line L4B2 fault and correct model judgment as follows: Figure 5 As shown, the measuring point P 11 With P 12 The connection between the measuring points P4 and P 11 and P 12 The weights at other measurement points are 0. At this time, the fault section range output by the model can be narrowed to the three measurement points. 11 The weight at is still the maximum among the three, indicating that the occurrence of topology reconstruction does not affect the correct judgment of the model. Through the above steps, the contribution of each measurement point information in the output result can be quantified, providing an important basis for improving the reliability of the output results of the fault location model.

[0110] Therefore, the present invention adopts the above-mentioned distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information. First, the feasibility of spatiotemporal graph information in the distribution network fault location scheme is analyzed, and a spatiotemporal graph information fault characterization structure integrating time continuity and spatial integrity is constructed; then, the time series waveform information and spatial structure characteristics of the distribution network fault are fully extracted through the spatiotemporal graph convolution network to achieve high-precision fault location in the scenario of frequent topological changes; finally, the spatiotemporal graph convolution network interpretability analysis module is constructed to perform post-interpretability analysis on the model decision basis and internal working mechanism. Through the above process, a closed-loop positioning scheme of "data-modeling-decision-verification" based on machine learning is formed, so that the positioning method has the advantages of high positioning accuracy and strong robustness, and maintains good generalization ability in complex operating scenarios such as new energy output fluctuations and topological changes, and has strong application value.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information, characterized in that: The following steps are involved: S1. Construct a fault representation structure of spatiotemporal graph information, integrate fault feature information from the perspective of temporal continuity and spatial integrity, and enhance the recognition ability of the classification model; S2. Establish a distribution network fault location model integrating spatiotemporal graph information; S3. Design an explainable analysis verification module to conduct explainable analysis on the decision-making basis and internal working mechanism.

2. According to claim 1, a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information is characterized by: In step S1, time continuity is to mine the electrical variation of fault information based on the continuous time series of voltage and current collected at the measuring point; Spatial integrity refers to the on-off change of the link relationship of topological nodes during the operation of the distribution network.

3. According to claim 1, a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information is characterized by: In step S1, the constructed spatiotemporal graph information fault representation structure expression is: Among them, p j is the measuring point numbered j in the distribution line; They represent the three-phase voltage and three-phase current collected at the measuring point at p j The instantaneous value of a moment.

4. According to claim 1, a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information is characterized by: In step S2, the fault location model includes a temporal information capture module and a spatial feature perception module; The timing information capture module uses the gated convolution unit as the main structure of the timing information capture module, and its mathematical expression is: Where: H l and H l+1 are the input and output of the Gated-CNN module at layer l; W1 l and is the convolution kernel of the same dimension; and is the bias of the model; is the convolution operation; σ1(·) is the gating function, which takes the Sigmoid activation function; The spatial feature perception module uses the distribution network topology structure based on the graph convolutional neural network to capture the impact of faults by aggregating node features. Its mathematical expression is: Where: X l ∈R N×C is the input matrix of the lth hidden layer, N is the number of measurement points, C is the number of sampling points of each measurement point; σ2(·) is the ReLU activation function; represents the sum of the adjacency matrix A and the identity matrix; Represents the degree matrix corresponding to the adjacency matrix A; W∈R C×Q , represents the weight matrix of the convolutional layer, Q is the number of hidden layer nodes, and finally the lth hidden layer output f(A,X l ).

5. According to claim 1, a distribution network fault section location and interpretability analysis method integrating spatiotemporal graph information is characterized by: In step S3, the interpretable analysis verification module combines graph data with gradient information to identify the feature nodes that the model focuses on in the decision-making process and evaluates the importance of graph nodes. The specific steps are as follows: S31, class activation loss update Assuming that the number of measurement points in the topology is n, the number of sampling channels of each measurement point is m, the sampling length is l, and the fault location task of c sections of the distribution network needs to be completed, the fault representation structure of the spatiotemporal graph information is X = {X1, X2, …, X n }, X∈R n ×m×l As the input of the interpretability stage, its label is Y = {Y1,Y2,…,Y n }; Among them, X i represents the timing signal collected at the i-th measurement point, Y i Indicates the fault section corresponding to the signal; Construct a new STGCN network with the same structure as the trained STGCN network, retain the training weight parameters, and define a new class activation loss function L CAM The expression is as follows: Among them, Z i ={Z1,Z2,Z3,…,Z n }, is X i The final classification output after inputting the model; the maximum value of the final classification output obtained by the argmax(·) function; With the defined class activation loss function L CAM As the loss function of STGCN, the model is trained so that the prediction result Z of each measurement point i As close to Y as possible i ; S32. Spatial weight map generation The fault representation structure X of the spatiotemporal graph information i Input L CAM The new STGCN network with loss function calculates the weight contribution of the sth measurement point to the final classification result The expression is as follows: Among them, d l+1 is the output dimension of the graph convolutional layer; y e Refers to the probability that the classification output is the e-th segment; The physical meaning of is the importance of the i-th feature point of the s-th measurement point to the classification result; By calculating, we can get the weight vector of each measurement point when the e-th section fails. The expression is as follows: According to the weight vector of each measurement point And the topological adjacency matrix A, the spatial weight of each node is used to judge the influence of node measurement information on the positioning result during the model classification process.

Images