Method and device for detecting short-circuit fault of power distribution network

Abstract

The invention discloses a method and device for detecting a short-circuit fault of a power distribution network, and relates to the technical field of power operation and maintenance, and the method comprises the steps: obtaining a topological structure of a target power distribution network and voltage and current data of each node during a fault period, constructing graph structure data, and adding a fault label; constructing a fault detection model comprising an LSTM network and a multi-layer GCN network, and introducing an adaptive frequency aggregation mechanism into each layer of GCN network; training the fault detection model through graph structure data with fault labels, and performing joint optimization on the fault detection model by utilizing a physical information constraint loss function and a classification loss function; and performing fault detection on the to-be-detected power distribution network through the trained fault detection model. Through joint modeling of time sequence and spatial features, the fault type and position can be accurately identified, the fault response time is effectively shortened, and the safety and stability of power grid operation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power operation and maintenance, and particularly to a method and device for detecting short - circuit faults in a distribution network. Background Art

[0002] As a core component of the power system, the distribution network undertakes the key task of delivering electric energy from the substation to the end - users. Due to the wide coverage, diverse equipment types, and complex operating environment of the distribution network, its reliability and stability directly affect the safety and economy of power supply. However, during the operation of the distribution network, various short - circuit faults (such as single - phase - to - ground short - circuit, two - phase - phase - to - phase short - circuit, three - phase - phase - to - phase short - circuit) can lead to voltage dips, abnormal currents, and even power supply interruptions, seriously affecting the power supply quality and user experience. Therefore, accurately and quickly detecting and locating distribution network faults is of great significance for ensuring the safety of the power system and reducing operation and maintenance costs.

[0003] Currently, distribution network fault detection technologies mainly include methods based on physical models, methods based on signal processing, and methods based on machine learning. Fault detection methods based on physical models include: impedance ranging method, traveling - wave method, fault - component method. The impedance ranging method and the fault - component method require accurate line impedance and system equivalent parameters. However, in the actual power grid, the parameters change greatly, making it difficult to accurately obtain. Methods such as the traveling - wave method require high - precision synchronous sampling equipment and have high computational complexity, making it difficult to achieve real - time detection. Methods based on signal processing include Fourier transform (FFT), wavelet transform (WT), and empirical mode decomposition (EMD). The limitations are that wavelet transform and empirical mode decomposition are sensitive to noise and are prone to misjudgment. Complex signal - processing algorithms require high computational resources and are not conducive to real - time applications. Their applicability is limited and the generality is poor. In methods based on machine learning, decision trees, support vector machines (SVMs), random forests, etc. have been introduced into the field of distribution network fault detection. These methods learn historical fault data and construct classification models for fault identification, but they are highly dependent on feature engineering and insufficient in processing time - series features, making it difficult to adapt to dynamic environments. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for detecting short - circuit faults in a distribution network, so as to solve the technical problems that traditional distribution network fault detection technologies have many deficiencies in the face of complex power grid structures, time - series feature extraction, and noise interference, and improve the accuracy and real - time performance of fault detection.

[0005] To achieve the above - mentioned purpose, the present invention is implemented by the following technical solutions:

[0006] In the first aspect, the present invention provides a method for detecting short - circuit faults in a distribution network, including:

[0007] Obtain the topological structure of the target distribution network and the voltage and current data of each node during the fault, construct graph structure data and add fault labels;

[0008] Construct a fault detection model that includes an LSTM network and a multi-layer GCN network, and introduce an adaptive frequency aggregation mechanism in each layer of the GCN network;

[0009] Train the fault detection model with the graph structure data with fault labels, and jointly optimize the fault detection model using a physical information constraint loss function and a classification loss function;

[0010] Perform fault detection on the distribution network to be detected through the trained fault detection model.

[0011] Optionally, the construction of the graph structure data with labels includes:

[0012] Construct an adjacency matrix according to the topological structure of the target distribution network , the row and column numbers of the adjacency matrix correspond to the nodes, and the of the adjacency matrix row column is :

[0013]

[0014] Construct a node feature matrix according to the voltage and current data of each node during the fault, including:

[0015] Preprocess the obtained voltage and current data, and construct a node feature matrix through the preprocessed voltage and current data , the row and column numbers of the node feature matrix correspond to the sampling time and the nodes, and the of the node feature matrix row column is :

[0016]

[0017] In the formula, is the preprocessed voltage and current data of node at the sampling time ;

[0018] Construct graph structure data according to the adjacency matrix and the node feature matrix .

[0019] Optionally, the preprocessing includes data cleaning and normalization processing.

[0020] Optionally, training the fault detection model with the graph structure data with fault labels and jointly optimizing the fault detection model using the physical information constraint loss function and the classification loss function includes:

[0021] Initializing the model parameters of the fault detection model;

[0022] Taking the graph structure data with fault labels as input data and inputting it into the fault detection model. Input the node feature matrix in the graph structure data into the LSTM network to extract temporal features, input the temporal features into the multi-layer GCN network, and use the adjacency matrix in the graph structure data to realize feature propagation between nodes to extract temporal features with spatial features, and perform mapping to obtain the fault classification probability of each node;

[0023] Calculating the physical information constraint loss function and the classification loss function based on the fault classification probability and fault labels of each node to obtain the total loss;

[0024] Updating the model parameters using the Adam optimizer based on the total loss, and combining the gradient clipping technique to limit the gradient norm;

[0025] Repeating the forward propagation and backward propagation processes until the total loss reaches a preset threshold or the number of training times reaches a set value, and completing the training of the fault detection model.

[0026] Optionally, introducing an adaptive frequency aggregation mechanism in each layer of the GCN network includes:

[0027] Using adaptive frequency aggregation to decompose the low-frequency component and high-frequency component in the GCN network, and dynamically adjusting the ratio of the low-frequency component and high-frequency component through learnable weights to enhance the adaptability of the GCN network to different node features. The expression of adaptive frequency aggregation is:

[0028]

[0029] In the formula, is the output feature of the GCN network, is the difference between the input feature of the GCN network and the feature , is the learnable weight, is the feature obtained by adaptive frequency aggregation and serves as the input feature of the next layer of the GCN network.

[0030] Optionally, the physical information constraint loss function includes:

[0031] Kirchhoff's current law loss :

[0032]

[0033] Wherein, is the total number of nodes, is the node set of neighbor nodes, is the node the current value flowing from the node to the node is the node the current value flowing from the node to the node

[0034] Kirchhoff's voltage law loss :

[0035]

[0036] Wherein, is the voltage value of the node ;

[0037] Optionally, the classification loss function is:

[0038]

[0039] Wherein, is the classification loss, is the total number of categories, is the category weight of, is the node fault label of, is the node predicted as the category probability of;

[0040]

[0041] Wherein, is the total number of nodes, is the category total number of nodes of, is the smoothing factor;

[0042] In a second aspect, the present invention provides a detection device for short - circuit faults in a distribution network, comprising:

[0043] A data processing module, configured to obtain the topological structure of the target distribution network and the voltage and current data of each node during a fault, construct graph - structured data and add fault labels;

[0044] A model construction module, configured to construct a fault detection model including an LSTM network and a multi - layer GCN network, and introduce an adaptive frequency aggregation mechanism into each layer of the GCN network;

[0045] A model training module, configured to train the fault detection model with graph structure data with fault labels, and jointly optimize the fault detection model by using a physical information constraint loss function and a classification loss function;

[0046] A model deployment module, configured to perform fault detection on a to-be-detected distribution network through the trained fault detection model.

[0047] In a third aspect, the present invention provides an electronic device, including a processor and a storage medium;

[0048] The storage medium is used to store instructions;

[0049] The processor is used to operate according to the instructions to execute the steps of the above method.

[0050] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0051] Compared with the prior art, the beneficial effects achieved by the present invention:

[0052] A method and device for detecting short-circuit faults in a distribution network provided by the present invention. 1) A fault detection model including an LSTM network and a multi-layer GCN network is constructed, and an adaptive frequency aggregation mechanism is introduced into each layer of the GCN network, so as to effectively improve the accuracy and reliability of fault detection by combining time series modeling and adaptive frequency aggregation. 2) The fault detection model is trained with graph structure data with fault labels, and the fault detection model is jointly optimized by using a physical information constraint loss function and a classification loss function; thereby introducing PINN physical constraints during the training process to ensure that the model conforms to the basic principles of the power system while performing data-driven learning, improving the generalization ability and physical interpretability. Compared with traditional fault detection methods based on rules or single machine learning models, the present invention can simultaneously consider the time evolution trend and spatial topology information of the distribution network, and improve the sensitivity and physical consistency to abnormal faults through AFA and PINN, so that the model still has strong robustness under a small amount of data or unseen fault patterns, and is particularly suitable for fault detection and location in an actual distribution network environment. Description of the Drawings

[0053] Figure 1 is a schematic flowchart of a method for detecting short-circuit faults in a distribution network provided by an embodiment of the present invention;

[0054] Figure 2 is a simulation wiring diagram of a distribution network with a three-microgrid structure provided by an embodiment of the present invention;

[0055] Figure 3It is a curve graph of the loss and accuracy during training provided by an embodiment of the present invention. Specific Embodiments

[0056] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0057] Embodiment 1:

[0058] As Figure 1 shown, an embodiment of the present invention provides a method for detecting short - circuit faults in a distribution network, including the following steps:

[0059] Step S1: Obtain the topological structure of the target distribution network and the voltage and current data of each node during the fault, construct graph - structured data, and add fault labels.

[0060] Specifically, in this embodiment, a three - microgrid - structured distribution network is selected as the target distribution network, and the simulation software MATLAB / SIMULINK is used for simulation. The specific simulation wiring diagram is as Figure 2 shown. Node 0 is the main bus node, nodes 1, 2, and 3 are the common load 3, 4, 5 - line nodes, node 4 is the bus 1 node, node 5 is the load bus 1 node, node 6 is the A - load - line node, nodes 7, 8, and 9 are the PV A1, A2, A3 - line nodes, node 10 is the battery - line 1 node, node 11 is the bus 2 node, node 12 is the load bus 2 node, node 13 is the B - load - line node, nodes 14, 15, and 16 are the PV B1, B2, B3 - line nodes, node 17 is the bus 3 node, node 18 is the load bus 3 node, node 19 is the C - load - line node, nodes 20, 21, 22, 23, and 24 are the PV automation 1, 2, 3, 4, 5 - line nodes, nodes 25 and 26 are the wind - turbine grid - connection 1, 2 - line nodes, node 27 is the battery - line 2 node, and node 28 is the load - line node.

[0061] In the simulation software, by manually setting the fault - component nodes, two nodes are randomly selected for each fault in the 29 nodes of the simulation model. The faults are set to 10 types of faults: any one of AG, BG, CG, ABG, ACG, BCG, AB, AC, BC, and ABC, which represent single - phase - to - ground faults of phases A, B, and C respectively; double - phase - to - ground faults of phases AB, AC, and BC respectively; double - phase short - circuit faults of phases AB, AC, and BC; and three - phase short - circuit fault of ABC. The simulation time is set to 1 s, and the fault occurs at 0.5 - 0.6 s. During the fault occurrence, the three - phase voltage and current data of each node are extracted, the sampling frequency is 500 HZ, and a total of 2900 groups of data are obtained. The ratio of fault - node data to non - fault - node data is 27:2. The simulation is run ten times repeatedly.

[0062] Specifically, in this embodiment, constructing the labeled graph structure data includes:

[0063] Construct an adjacency matrix according to the topological structure of the target distribution network , where the row and column numbers of the adjacency matrix correspond to the nodes, and the of the adjacency matrix row and column are:

[0064]

[0065] Construct a node feature matrix according to the voltage and current data of each node during the fault, including:

[0066] Preprocess the obtained voltage and current data, and construct a node feature matrix through the preprocessed voltage and current data , where the row and column numbers of the node feature matrix correspond to the sampling time and the nodes, and the of the node feature matrix row and column are:

[0067]

[0068] In the formula, is the preprocessed voltage and current data of node at the sampling time ;

[0069] Construct the graph structure data according to the adjacency matrix and the node feature matrix .

[0070] Among them, the preprocessing includes data cleaning and normalization processing.

[0071] Data cleaning, such as removing noise and outliers, filling in missing values, can provide high-quality and reliable input data for the model, thereby improving the training efficiency and prediction accuracy.

[0072] The voltage and current data of each node collected are extremely unbalanced during faults and non-faults. At the same time, there are differences in the dimensions of current and voltage, and there is a large gap in amplitude. The different dimensions and amplitude gaps will have a greater impact on model training, resulting in a decline in the model training effect. Therefore, it is necessary to normalize the data.

[0073] Step S2: Construct a fault detection model that includes an LSTM network and a multi-layer GCN network, and introduce an adaptive frequency aggregation mechanism into each layer of the GCN network.

[0074] The LSTM (Long Short-Term Memory) network is a long short-term memory network, which is a special structure of recurrent neural network. Compared with the traditional BP neural network, the training parameters of the LSTM network neurons increase the long and short-term memory data, and can better capture the information of time series change data.

[0075] The GCN (Graph Convolution Networks) network is a graph convolution network, which is a special convolutional neural network structure for graph data processing. Similar to the convolutional neural network for processing image data, the graph convolutional neural network extracts features from the input graph data through convolutional operations for tasks such as node classification, graph classification, and link prediction. The GCN network mainly includes an input layer and a graph convolutional layer. The graph convolutional layer is the core component, which updates the representation of the current node by aggregating the features of neighboring nodes. The core idea is the neighborhood aggregation of node features, that is, the features of each node are not only determined by its own features, but also affected by the features of its neighboring nodes.

[0076] Through the joint learning of GCN and LSTM, LSTM-GCN can simultaneously consider spatial (node topology) and temporal (temporal variation) information, providing more accurate fault classification and fault location capabilities. LSTM processes temporal features, captures the changing laws of current and voltage over time, models long-term dependencies, and avoids the misclassification problems caused by traditional methods based only on single time-step data. GCN processes the power grid topology information, models using the adjacency relationship of the distribution network, and realizes feature propagation between nodes, so that the fault detection not only depends on the data of a single node, but also combines the information of adjacent nodes, improving the detection accuracy. The two neural network modules are combined sequentially. The input of the LSTM network is the time series features of the nodes, and the output is the hidden layer features, which are used to capture the time dependencies of the node features; the input of the GCN network is the output features of the LSTM network. Each layer of the GCN network performs neighbor feature aggregation (message passing), allowing each node to fuse the information from adjacent nodes and enhancing the perception of the global network state. After multiple graph convolutional operations, the fault classification probability of each node is output.

[0077] Step S3: Train the fault detection model with graph structure data with fault labels, and jointly optimize the fault detection model using the physical information constraint loss function and the classification loss function.

[0078] Specifically, in this embodiment, the training process is as follows:

[0079] Step S3.1: Initialize the model parameters of the fault detection model; the model parameters include the network parameters of the LSTM network, the network parameters of the multi-layer GCN network, and the weight parameters of the adaptive frequency aggregation mechanism.

[0080] Step S3.2: Input the graph structure data with fault labels as input data into the fault detection model. Input the node feature matrix in the graph structure data into the LSTM network to extract temporal features, input the temporal features into the multi-layer GCN network, and use the adjacency matrix in the graph structure data to realize feature propagation between nodes to extract temporal features with spatial features, and perform mapping to obtain the fault classification probabilities of each node.

[0081] Step S3.3: Calculate the physical information constraint loss function and the classification loss function based on the fault classification probabilities and fault labels of each node to obtain the total loss.

[0082] The physical information constraint loss function includes:

[0083] Kirchhoff's current law loss :

[0084]

[0085] In the formula, is the total number of nodes, is the node set of neighbor nodes of, is the node current value flowing from the node to the node , is the node current value flowing from the node to the node ;

[0086] Kirchhoff's voltage law loss :

[0087]

[0088] In the formula, is the voltage value of the node .

[0089] The classification loss function is:

[0090]

[0091] In the formula, is the classification loss, is the total number of categories, is the category weight of, is the node fault label of, is the node The probability of being predicted as a category ;

[0092]

[0093] In the formula, is the total number of nodes, is the total number of nodes of category ; is the smoothing factor. The smoothing factor is used to prevent the weights from being too large or too small. Generally, it takes a value of 0.1 - 0.5. The larger its value, the more significant the impact of class imbalance.

[0094] The total loss is:

[0095]

[0096] In the formula, is the Kirchhoff's current law loss and Kirchhoff's voltage law loss, is 's weight.

[0097] In the training process of the present invention, PINN physical constraints are introduced, and the loss function is constructed based on Kirchhoff's current law (KCL) and Kirchhoff's voltage law (KVL). Combining with the classification loss function ensures that the model conforms to the basic principles of the power system while performing data-driven learning, improving the generalization ability and physical interpretability.

[0098] Step S3.4: Update the model parameters using the Adam optimizer based on the total loss, and limit the gradient norm by combining with the gradient clipping technique.

[0099] Step S3.5: Repeat the forward propagation and backward propagation processes until the total loss reaches the preset threshold or the number of training times reaches the set value, and complete the training of the fault detection model.

[0100] Most traditional GCN networks extract common information from nodes and retain the commonalities of nodes, which can be regarded as low-pass filters. However, when performing fault detection, low-pass filtering will cause the differences between nodes to be ignored. The differential features between fault nodes and non-fault nodes should be maintained, so that the node embeddings of fault nodes and non-fault nodes have a high degree of distinguishability. Therefore, the low-frequency components and high-frequency components in the GCN network are decomposed using adaptive frequency aggregation, and their proportions are dynamically adjusted through learnable weights to enhance the adaptability of the model to different types of node features. The expression of adaptive frequency aggregation is:

[0101]

[0102] In the formula, is the output feature (low-frequency information) of the GCN network, which is obtained by the convolutional calculation of the GCN network and represents the smoothed information propagated from neighbor nodes. It is suitable for global pattern recognition (such as group faults). is the difference (high-frequency information) between the input feature and the feature of the GCN network, which retains local details (such as outliers and edge node features). is the learnable weight. is the feature obtained by adaptive frequency aggregation and serves as the input feature for the next layer of the GCN network.

[0103] Step S4: Use the trained fault detection model to detect faults in the power distribution network to be detected.

[0104] During the deployment and use process, input the test data into the trained model to output the fault classification results of each node; count the fault nodes and their fault types at each time step; output the fault node and fault type information at each time step to assist the operation and maintenance personnel in quickly locating faults. This method supports real-time data processing and inputs the running data and outputs the fault detection results in JSON format.

[0105] Combined with the data obtained from the above simulation for training and verification, the machine learning framework is Pytorch. The learning rate used during training is 0.01, the training epoch is set to 200, and the physical information constraint loss weight is set to 0.2. Train the data. The loss and accuracy during training are as Figure 3 shown. As shown in Table 1, the accuracy of the proposed improved GCN-LSTM model on the 127 laboratory power grid simulation dataset can reach 96.8%. Compared with other deep learning models, the model proposed in the present invention shows extremely high superiority.

[0106] Table 1. Comparison of GCN-LSTM model with other models

[0107]

[0108] At the same time, in order to study the influence of noise on model analysis, noises with signal-to-noise ratios of 5% and 10% are added to the simulation data respectively, and training and classification are carried out again. The accuracy of the GCN-LSTM model under noise is shown in Table 2.

[0109] Table 2. Detection accuracy of GCN-LSTM model under noise

[0110]

[0111] As shown in Table 2, this model has a certain anti-interference ability against noise. When there is 10% noise interference, the model accuracy only drops by 4%. The influence of noise interference with a signal-to-noise ratio of 5% on the accuracy is even more negligible, showing the strong robustness of the model.

[0112] In summary, the embodiments of the present invention have the following characteristics:

[0113] (1) The spatio-temporal feature extraction ability combined by LSTM-GCN. The LSTM layer can effectively capture the time-series features of the node voltage and current in the distribution network, and identify the dynamic change patterns of fault occurrence. By processing multi-time-step data, LSTM captures the time dependence and periodic features, improving the detection ability for sudden faults and gradually evolving faults. The GCN layer, based on the topological structure of the distribution network, extracts the spatial relationship features between nodes. GCN propagates feature information through the adjacency matrix, making full use of the physical connection characteristics of the power grid, and can accurately locate the fault propagation path and influence range. The combination of the two can achieve the spatio-temporal joint detection of distribution network faults, significantly improving the accuracy and robustness of fault identification.

[0114] (2) The adaptive frequency aggregation (AFA) mechanism improves the quality of feature extraction. After each layer of GCN convolution, the AFA mechanism automatically distinguishes and adjusts the high-frequency information (abnormal signals, local features) and low-frequency information (smooth features, global patterns). Through learnable parameters, the model adaptively determines the weights of high- and low-frequency features, avoiding the loss of feature information caused by over-smoothing in traditional models. It enhances the sensitivity of the model to the details of fault signals. Especially when detecting complex or weak fault signals, it can better retain key features, helping to improve the fault location accuracy of the model.

[0115] (3) Introducing physical information constraints to enhance the physical consistency of the model. During the model training process, physical constraints based on Kirchhoff's current law (KCL) and Kirchhoff's voltage law (KVL) of the power system are added. These physical laws ensure the conservation relationship of node currents and voltages, making the model conform to the basic physical laws of the distribution network during prediction. This enables the model to not only rely on data-driven but also ensure the physical interpretability and consistency of the prediction results, reducing non-physical results caused by overfitting, and significantly improving the reliability of the model. It reduces the dependence on large-scale labeled data. After introducing physical constraints, the model can still maintain good performance with a small amount of data by following the physical laws of the power system. This method is particularly important in practical application scenarios with scarce data, which can reduce the data acquisition cost and improve the generalization ability and practicality of the model.

[0116] Embodiment 2:

[0117] The embodiment of the present invention provides a detection device for short-circuit faults in a distribution network, including:

[0118] A data processing module, configured to obtain the topological structure of the target distribution network and the voltage and current data of each node during a fault, construct graph structure data and add fault labels;

[0119] A model construction module, configured to construct a fault detection model including an LSTM network and a multi-layer GCN network, and introduce an adaptive frequency aggregation mechanism into each layer of the GCN network;

[0120] A model training module, configured to train the fault detection model through the graph structure data with fault labels, and jointly optimize the fault detection model by using a physical information constraint loss function and a classification loss function;

[0121] A model deployment module, configured to perform fault detection on the distribution network to be detected through the trained fault detection model.

[0122] Embodiment III:

[0123] Based on the detection method provided in Embodiment I, an embodiment of the present invention provides an electronic device, including a processor and a storage medium;

[0124] The storage medium is used to store instructions;

[0125] The processor is used to operate according to the instructions to execute the steps of the above method.

[0126] Embodiment IV:

[0127] Based on the detection method provided in Embodiment I, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0128] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0130] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in the Figure 1 one or more flows and / or blocks Figure 1 one or more blocks.

[0132] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principles of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for detecting a short-circuit fault in a distribution network, characterized in that: include: Obtain the topological structure of the target distribution network and the voltage and current data of each node during the fault period, construct graph structure data and add fault labels; Build a fault detection model consisting of LSTM network and multi-layer GCN network, and introduce an adaptive frequency aggregation mechanism in each layer of GCN network; The fault detection model is trained by using graph structure data with fault labels, and the fault detection model is jointly optimized by using a physical information constraint loss function and a classification loss function; The fault detection model is used to perform fault detection on the distribution network to be tested.

2. The method for detecting a short-circuit fault in a distribution network according to claim 1, characterized in that: The step of constructing the labeled graph structure data includes: Construct an adjacency matrix based on the topological structure of the target distribution network , the adjacency matrix The row and column numbers correspond to the nodes, and the adjacency matrix of OK Listed as : ； A node characteristic matrix is ​​constructed according to the voltage and current data of each node during the fault period, including: Preprocess the acquired voltage and current data, and construct a node feature matrix based on the preprocessed voltage and current data , node feature matrix The row and column numbers of the node correspond to the sampling time and node, and the node feature matrix of OK Listed as : ； In the formula, For Node At sampling time The pre-processed voltage and current data; According to the adjacency matrix and the node feature matrix Building graph structured data .

3. The method for detecting a short-circuit fault in a distribution network according to claim 2, characterized in that: The preprocessing includes data cleaning and normalization.

4. The method for detecting a short-circuit fault in a distribution network according to claim 1, characterized in that: The method of training the fault detection model by using graph structure data with fault labels and jointly optimizing the fault detection model by using a physical information constraint loss function and a classification loss function includes: Initialize model parameters of the fault detection model; Input the graph structure data with fault labels as input data into the fault detection model, input the node feature matrix in the graph structure data into the LSTM network to extract the time series features, input the time series features into the multi-layer GCN network, and use the adjacency matrix in the graph structure data to realize the feature propagation between nodes to extract the time series features with spatial features, and perform mapping to obtain the fault classification probability of each node; The total loss is obtained by calculating the physical information constraint loss function and the classification loss function according to the fault classification probability and fault label of each node; Based on the total loss, the Adam optimizer is used to update the model parameters, and the gradient norm is limited in combination with the gradient clipping technique; Repeat the forward propagation and back propagation processes until the total loss reaches a preset threshold or the number of training times reaches a set value, completing the training of the fault detection model.

5. The method for detecting a short-circuit fault in a distribution network according to claim 1, characterized in that: The introduction of an adaptive frequency aggregation mechanism in each layer of the GCN network includes: Adaptive frequency aggregation is used to decompose the low-frequency components and high-frequency components in the GCN network, and the ratio of low-frequency components to high-frequency components is dynamically adjusted through learnable weights to enhance the adaptability of the GCN network to different node characteristics. The expression of adaptive frequency aggregation is: ； In the formula, Output features for the GCN network, Input features and characteristics for GCN network The difference, are learnable weights, The features obtained by adaptive frequency aggregation are used as the input features of the next layer of GCN network.

6. The method for detecting a short-circuit fault in a distribution network according to claim 1, characterized in that: The physical information constraint loss function includes: Kirchhoff's Current Law Losses : ； In the formula, is the total number of nodes, For Node The set of neighbor nodes of For Node Flow to Node The current value, For Node Flow to Node The current value; Kirchhoff's Voltage Law Losses : ； In the formula, For Node Voltage value.

7. The method for detecting a short-circuit fault in a distribution network according to claim 1, characterized in that: The classification loss function is: ； In the formula, is the classification loss, is the total number of categories, For Category The weight of For Node The fault label, For Node Predicted as category probability; ； In the formula, is the total number of nodes, For Category The total number of nodes, is the smoothing factor.

8. A detection device for short-circuit fault in a distribution network, characterized in that: include: A data processing module is configured to obtain the topological structure of the target distribution network and the voltage and current data of each node during the fault period, construct graph structure data and add fault labels; A model building module is configured to build a fault detection model including an LSTM network and a multi-layer GCN network, and introduce an adaptive frequency aggregation mechanism in each layer of the GCN network; A model training module is configured to train the fault detection model using graph structure data with fault labels, and jointly optimize the fault detection model using a physical information constraint loss function and a classification loss function; The model deployment module is configured to perform fault detection on the distribution network to be detected by using the trained fault detection model.

9. An electronic device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images