Power distribution network fault prediction method, model training method, medium and equipment

Through variational modal decomposition and mutual information entropy analysis technology combined with bidirectional gated cyclic units and graph convolutional network, a distribution network fault model is constructed, which solves the problems of unstable fault detection accuracy and slow processing speed in the existing technology. It achieves more efficient feature extraction and fault detection effects.

CN119090049APending Publication Date: 2024-12-06YUNNAN POWER GRID CO LTD ELECTRIC POWER RES INST
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411080985.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing deep learning models have unstable accuracy and slow processing speed in distribution network fault detection, making it difficult to effectively capture temporal and spatial characteristics.

Method used

Variable modal decomposition and mutual information entropy analysis technology decompose sample fault data, build a parallel bidirectional gating cyclic unit and graph convolutional network model, feature reconstruction through the full connection layer, spatial and temporal features are extracted, and spatial and temporal features are extracted. Removal of redundant information through the channel attention mechanism.

Benefits of technology

It improves the accuracy and efficiency of fault detection, can capture time and space characteristics at the same time, and improves the model's fault detection capabilities and processing speed.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090049A_ABST
    Figure CN119090049A_ABST
Patent Text Reader

Abstract

The invention discloses a power distribution network fault prediction method, a model training method, a medium and equipment, and the method comprises the steps: firstly collecting the sample fault data of a power distribution network, and carrying out the decomposition through employing a variational mode decomposition technology, and obtaining a plurality of intrinsic mode function components; performing mutual information entropy analysis on the components to obtain an adjacent matrix and a feature matrix, and further constructing a graph convolutional network; the power distribution network fault model comprises a bidirectional gating circulation unit and a graph convolution network which are parallel, and a full connection layer connecting the bidirectional gating circulation unit and the graph convolution network. And inputting the first eigenmode function component and the adjacent matrix into a graph convolutional network to extract spatial features, and inputting the second eigenmode function component into a bidirectional gating loop unit to extract time features. And then, inputting the extracted spatial features and time features into a full connection layer for feature reconstruction, outputting a fault prediction result, and performing iterative training until the model converges. The model can capture time and space features at the same time and reconstruct the time and space features, and the accuracy and efficiency are effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault diagnosis, and in particular to a distribution network fault prediction method, a model training method, a medium and a device. Background Art

[0002] Nowadays, deep learning technology has been rapidly applied to the field of power grid fault diagnosis. Commonly used deep learning models include convolutional neural networks, recurrent neural networks, and deep belief networks. Recently, graph convolutional networks have also been applied to the field of power grid fault diagnosis because they can better simulate distribution networks. However, this deep learning model still has problems such as unstable fault detection accuracy and slow processing speed. Summary of the invention

[0003] Based on this, it is necessary to provide distribution network fault prediction methods, model training methods, media and equipment to solve the problems of unstable fault detection accuracy and slow processing speed.

[0004] A method for training a distribution network fault model, the method comprising:

[0005] Acquire sample fault data of the distribution network; wherein the sample fault data carries a corresponding fault label;

[0006] Decomposing the sample fault data by using variational mode decomposition to obtain a plurality of eigenmode function components of the sample fault data;

[0007] Performing mutual information entropy analysis on multiple eigenmode function components of sample fault data to obtain an adjacency matrix and a feature matrix, and constructing a graph convolution network based on the adjacency matrix and the feature matrix;

[0008] Constructing a distribution network fault model; wherein the distribution network fault model includes a parallel bidirectional gated recurrent unit and a graph convolutional network, and a fully connected layer connecting the bidirectional gated recurrent unit and the graph convolutional network;

[0009] Inputting the first eigenmode function component and the adjacency matrix into a graph convolutional network for feature extraction to extract spatial features, and inputting the second eigenmode function component into a bidirectional gated recurrent unit for feature extraction to extract temporal features; wherein the first eigenmode function component and the second eigenmode function component are any one of a plurality of eigenmode function components of sample fault data;

[0010] The spatial features and the temporal features are input into the fully connected layer for reconstruction, and the output fault prediction results are obtained. The distribution network fault model is cyclically trained according to the fault prediction results and the fault labels until the distribution network fault model converges.

[0011] In one embodiment, the applying variational mode decomposition to decompose the sample fault data to obtain a plurality of eigenmode function components of the sample fault data includes:

[0012] For the sample fault data, different modes are set to have corresponding center frequencies and bandwidths, and a constraint model is set; wherein the constraint model is used to minimize the sum of the estimated bandwidths of each mode, and the constraint model is expressed as:

[0013]

[0014] In the above formula, K is the number of preset modes, {u k} is the kth mode component after decomposition, {ω k} is the kth center frequency after decomposition, δ(t) is the Dirac function;

[0015] Based on the introduced penalty factor and Lagrange multiplication operator, the constrained model is converted into an unconstrained model, and the unconstrained model is expressed as:

[0016]

[0017] In the above formula, α is the penalty factor and λ is the Lagrange multiplication operator;

[0018] Iteratively update the parameter u by alternating direction of multipliers k ,ω k and λ, expressed as:

[0019]

[0020]

[0021] In the above formula, and represents f(ω), u i (ω), λ(ω), and Fourier transform of , n is the number of iterations, γ is the noise tolerance;

[0022] like Then return to execute the iterative update parameter u through the multiplier alternating direction method k ,ω k and λ steps and subsequent steps until To obtain K decomposed eigenmode function components.

[0023] In one embodiment, the step of applying variational mode decomposition to decompose the sample fault data to obtain a plurality of eigenmode function components of the sample fault data further includes:

[0024] Mutual information entropy is applied to perform correlation analysis on the intrinsic mode function components with the sample fault data as the target value to obtain multiple correlation values, which are expressed as:

[0025]

[0026] MI(IMF,X(t))=H(X(t))-H(X(t)|IMF)

[0027] In the above formula, i = 1, 2, ..., K, mi represents the i-th correlation value, H(X(t)) represents the entropy of the sample fault data X(t), and H(X(t)|IMF) is the conditional entropy of the sample fault data X(t) when the intrinsic mode function component IMF is known;

[0028] A preset number K′ of eigenmode function components with the highest correlation values ​​are retained; wherein K′<K.

[0029] In one embodiment, the mutual information entropy analysis is performed on multiple eigenmode function components of the sample fault data to obtain an adjacency matrix and a feature matrix, including:

[0030] Based on the mutual information entropy value between two nodes in each eigenmode function component, a K′×K′ connection matrix is ​​constructed as an adjacency matrix, where the mutual information entropy value is a measure of the degree of mutual dependence between two random variables;

[0031] The absolute value of the normalized mutual information entropy between the eigenmode function components is used as the spatial feature of the adjacent components to construct a K′×F-dimensional feature matrix; where F is the length of the historical time series of the component.

[0032] In one embodiment, the information transfer function of the graph convolutional network is expressed as:

[0033]

[0034] In the above formula, σ represents the Sigmoid activation function, H (l) represents the network output of layer l, W (l) represents the parameters to be trained, for The degree matrix, A represents the adjacency matrix, I N represents the N-order identity matrix, Represents the processed adjacency matrix.

[0035] In one embodiment, the bidirectional gated recurrent unit model at time t is expressed as:

[0036] r t =σ(W r ·[h t-1 ,xt ])

[0037] z t =σ(W z ·[h t-1 ,x t ])

[0038]

[0039] In the above formula, r t is the reset gate, z t is the update gate, x t is the current input, is the current input and the previous hidden layer state summary, h t is the hidden layer output after updating the memory, W r is the weight matrix of the reset gate, W z is the weight matrix of the update gate, W h is the weight matrix of the hidden layer, σ is the Sigmoid function, and tanh is the hyperbolic tangent activation function.

[0040] In one embodiment, the distribution network fault model further includes a channel attention mechanism layer, which is arranged between the bidirectional gated recurrent unit and the fully connected layer, and between the graph convolutional network and the fully connected layer. The method includes: applying the channel attention mechanism layer to remove the redundant information contained in the extracted spatial features and the temporal features, which is specifically expressed as follows:

[0041] Q=W Q *x i

[0042] K=W K *x i

[0043] V=W V *x i

[0044]

[0045] out i =sum(A i *V j ),j=1,2,...,n

[0046] Among them, W Q Represents the weight matrix of the linear transformation of the query, W K Represents the weight matrix for linear transformation of keys, W V Represents the weight matrix for linear transformation of values, x i represents the i-th feature of the input, d K represents the dimension of the key vector, KT Indicates matrix transposition, out i Represents the i-th element of the output sequence, V j Represents the jth element of the value sequence, A i represents the attention weight of the i-th element.

[0047] A method for predicting distribution network faults, the method comprising:

[0048] Obtain current fault data of the distribution network;

[0049] Decomposing the current fault data by applying variational mode decomposition to obtain a plurality of eigenmode function components of the current fault data;

[0050] Input multiple eigenmode function components of current fault data into the distribution network fault model for prediction to obtain the current fault prediction result; wherein, the distribution network fault model is trained by any one of the methods according to claims 1-7.

[0051] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the processor executes the steps of a distribution network fault model training method and a distribution network fault prediction method.

[0052] A terminal device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of a distribution network fault model training method and a distribution network fault prediction method.

[0053] The present invention provides a distribution network fault prediction method, a model training method, a medium and a device. Sample fault data of the distribution network is first collected, and then the sample fault data is decomposed by applying the variational mode decomposition technology to obtain a plurality of eigenmode function components. Subsequently, mutual information entropy analysis is performed on these components to obtain an adjacency matrix and a feature matrix, and a graph convolution network is constructed based on these matrices. The constructed distribution network fault model includes a parallel bidirectional gated cyclic unit and a graph convolution network, and a fully connected layer connecting the two. The first eigenmode function component and the adjacency matrix are input into the graph convolution network to extract spatial features, and the second eigenmode function component is input into the bidirectional gated cyclic unit to extract temporal features. Then, the extracted spatial features and temporal features are input into the fully connected layer for feature reconstruction, and the fault prediction results are output, and the model is cyclically trained according to the prediction results and the fault labels until the model converges. In this embodiment, variational mode decomposition and mutual information entropy analysis make the feature extraction process more efficient and accurate. At the same time, by using bidirectional gated recurrent units and graph convolutional networks in parallel, the distribution network fault model can simultaneously capture and reconstruct temporal and spatial features, which improves the accuracy and efficiency of model fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] in:

[0056] Figure 1 A schematic diagram of a flow chart of a method for training a distribution network fault model in one embodiment;

[0057] Figure 2 A schematic diagram of a flow chart of a method for predicting a distribution network fault in an embodiment;

[0058] Figure 3 It is a schematic diagram of the IEEE 33 node distribution network model;

[0059] Figure 4 This is a structural block diagram of the terminal equipment. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0062] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0063] like Figure 1 As shown, Figure 1 FIG. 1 is a flow chart of a method for training a distribution network fault model in an embodiment. The method for training a distribution network fault model in this embodiment includes the following steps:

[0064] S101, obtaining sample fault data of the distribution network.

[0065] Among them, sample fault data refers to the electrical data collected from the wind power distribution network system when a fault occurs. The data usually includes parameters such as voltage and current before and after the fault, and these sample fault data carry corresponding fault labels, such as short circuit fault, switch fault, etc.

[0066] Optionally, we use the improved IEEE 33 simulation model to generate sample fault data through simulation software (MATLAB / Simu link). The data should include current and voltage data and zero-sequence voltage for one week before and after the fault. Then mark the fault data, for example, mark short circuit faults, switch faults, etc.

[0067] S102, applying variational mode decomposition to decompose the sample fault data to obtain multiple eigenmode function components of the sample fault data.

[0068] Each eigenmode function component I MF is a sub-signal containing a specific frequency feature, having the same time length but different frequencies and amplitudes.

[0069] In a specific embodiment, decomposing to obtain the eigenmode function components specifically comprises the following steps:

[0070] A1. For sample fault data, set different modes with corresponding center frequencies and bandwidths, and set constraint models.

[0071] Among them, the center frequency refers to the main frequency of each mode component, and the bandwidth refers to the frequency range covered by the mode component. The center frequency and bandwidth are set to determine the position and range of each mode component in the frequency domain. The constraint model is used to minimize the sum of the estimated bandwidths of each mode, thereby ensuring that the decomposed mode components have a good frequency separation effect. The constraint model is expressed as:

[0072]

[0073] In the above formula, K is the number of preset modes, {u k} is the kth mode component after decomposition, {ω k} is the kth center frequency after decomposition, δ(t) is the Dirac function;

[0074] A2. Based on the introduced penalty factor and Lagrange multiplication operator, the constrained model is converted into an unconstrained model. The unconstrained model is expressed as:

[0075]

[0076] In the above formula, α is the penalty factor and λ is the Lagrange multiplication operator; the penalty factor is used to balance the weights of the constraint terms and the data fitting terms, and the Lagrange multiplication operator is used to convert the constraint conditions into additional terms in the optimization problem, so that the optimization problem can be solved by the unconstrained optimization method.

[0077] A3. Iteratively update the parameter u by the multiplier alternating direction method k ,ω k and λ.

[0078] It is expressed as:

[0079]

[0080] In the above formula, and represents f(ω), u i (ω), λ(ω), and The Fourier transform of , n is the number of iterations, γ is the noise tolerance; the multiplier gradually approaches the optimal solution by alternately updating each variable. Here it is used to iteratively update the parameters of each mode component until the convergence condition is met.

[0081] A4. Determine whether

[0082] like Then return to execute A3 and subsequent steps until Execute A5 to obtain K decomposed eigenmode function components.

[0083] Optionally, in order to construct the model more effectively later, after applying variational mode decomposition to decompose the sample fault data to obtain multiple eigenmode function components of the sample fault data, the following steps may be performed:

[0084] A6. Apply mutual information entropy and perform correlation analysis on the intrinsic mode function components with sample fault data as the target value to obtain multiple correlation values, which are expressed as:

[0085]

[0086] MI(IMF,X(t))=H(X(t))-H(X(t)|IMF)

[0087] In the above formula, i = 1, 2, ..., K, mi represents the i-th correlation value, H(X(t)) represents the entropy of the sample fault data X(t), and H(X(t)|IMF) is the conditional entropy of the sample fault data X(t) when the intrinsic mode function component IMF is known. These correlation values ​​represent the contribution of each intrinsic mode function component to the sample fault data. The larger the correlation value, the higher the contribution.

[0088] A7. Retain a preset number K′ of eigenmode function components with the highest correlation values.

[0089] Among them, K′<K. Therefore, through the above steps, the eigenmode function components of the sample fault data can be screened to ensure that the components used in the subsequent analysis have high representativeness and relevance. After screening out the highly relevant components, the model can be constructed more effectively, reducing the computational burden and improving the training and prediction efficiency of the model.

[0090] S103, performing mutual information entropy analysis on multiple eigenmode function components of the sample fault data to obtain an adjacency matrix and a feature matrix, and constructing a graph convolutional network based on the adjacency matrix and the feature matrix.

[0091] In a specific embodiment, performing mutual information entropy analysis on multiple eigenmode function components of sample fault data to obtain an adjacency matrix and a feature matrix specifically includes the following steps:

[0092] B1. Construct a K′×K′ connection matrix as an adjacency matrix based on the mutual information entropy value between two nodes in each eigenmode function component.

[0093] First, the mutual information entropy value is calculated. The mutual information entropy value is a measure of the degree of mutual dependence between two random variables (two nodes in each eigenmode function component). Then, the mutual information entropy value is used to construct an adjacency matrix to represent the connection relationship between these components.

[0094] B2. The absolute value of the normalized mutual information entropy between the eigenmode function components is used as the spatial feature of the adjacent components to construct a K′×F dimensional feature matrix.

[0095] Where F is the length of the historical time series of the component.

[0096] Next, a graph convolutional network (X, A) is constructed based on the adjacency matrix and feature matrix. The information transfer function of the graph convolutional network is expressed as:

[0097]

[0098] In the above formula, σ represents the Sigmoid activation function, H (l) represents the network output of layer l, W (l) represents the parameters to be trained, for The degree matrix, A represents the adjacency matrix, I N represents the N-order identity matrix, Represents the processed adjacency matrix.

[0099] S104, construct a distribution network fault model.

[0100] The distribution network fault model includes a parallel bidirectional gated recurrent unit and a graph convolutional network, and a fully connected layer connecting the bidirectional gated recurrent unit and the graph convolutional network. The bidirectional gated recurrent unit is used to process time series data. It processes sequence data in both the forward and reverse directions and can capture the dependencies between the sequences. The graph convolutional network is a neural network that processes graph structured data. It aggregates the feature information of the node neighborhood through graph convolution operations and learns the node feature representation in the graph structure.

[0101] Optional, bidirectional gated recurrent unit model at time t, expressed as:

[0102] r t =σ(W r ·[h t-1 ,x t ])

[0103] z t =σ(W z ·[h t-1 ,x t ])

[0104]

[0105] In the above formula, r t is the reset gate, z t is the update gate, x t is the current input, is the current input and the previous hidden layer state summary, h t is the hidden layer output after updating the memory, W r is the weight matrix of the reset gate, W z is the weight matrix of the update gate, W h is the weight matrix of the hidden layer, σ is the Sigmoid function, and tanh is the hyperbolic tangent activation function.

[0106] S105, input the first eigenmode function component and the adjacency matrix into the graph convolutional network for feature extraction to extract spatial features, and input the second eigenmode function component into the bidirectional gated recurrent unit for feature extraction to extract temporal features.

[0107] The first eigenmode function component and the second eigenmode function component are any one of the multiple eigenmode function components of the sample fault data. In other words, the bidirectional gated recurrent unit and the graph convolutional network process the eigenmode function components in parallel, which can simultaneously capture the time-dependent characteristics and spatial topological characteristics in the distribution network fault data, thereby improving the accuracy of fault detection and also improving efficiency.

[0108] Optionally, the distribution network fault model further includes a channel attention mechanism layer, which is arranged between the bidirectional gated recurrent unit and the fully connected layer, and between the graph convolutional network and the fully connected layer. The following steps may also be performed: applying the channel attention mechanism layer to remove redundant information contained in the extracted spatial features and temporal features, which is specifically expressed as follows:

[0109] Q=W Q *x i

[0110] K=W K *x i

[0111] V=W V *x i

[0112]

[0113] out i =sum(A i *V j ),j=1,2,...,n

[0114] Among them, W Q Represents the weight matrix of the linear transformation of the query, W K Represents the weight matrix for linear transformation of keys, W V The weight matrix representing the linear transformation of the value, x i represents the i-th feature of the input, d K represents the dimension of the key vector, K T Indicates matrix transposition, out i Represents the i-th element of the output sequence, V j Represents the jth element of the value sequence, A i represents the attention weight of the i-th element.

[0115] In this way, through the channel attention mechanism layer, the model can remove redundant information, retain important features, and improve the accuracy of feature representation.

[0116] S106, input the spatial features and the temporal features into the fully connected layer for reconstruction, obtain the output fault prediction results, and cyclically train the distribution network fault model according to the fault prediction results and the fault labels until the distribution network fault model converges.

[0117] Specifically, the fully connected layer processes the input features comprehensively and outputs the reconstructed feature representation. Based on these feature representations, the model generates fault prediction results for the distribution network. The fault prediction results are then compared with the actual fault labels to calculate the loss function. Based on the value of the loss function, the model parameters are adjusted using the gradient descent method or other optimization algorithms. The above process is repeated and the model is trained cyclically until the loss function of the model reaches a stable minimum value, that is, the model converges.

[0118] The training method of the distribution network fault model is to first collect sample fault data of the distribution network, and then apply the variational mode decomposition technology to decompose the sample fault data to obtain multiple eigenmode function components. Subsequently, the components are subjected to mutual information entropy analysis to obtain an adjacency matrix and a feature matrix, and a graph convolution network is constructed based on these matrices. The constructed distribution network fault model includes a parallel bidirectional gated cyclic unit and a graph convolution network, and a fully connected layer connecting the two. The first eigenmode function component and the adjacency matrix are input into the graph convolution network to extract spatial features, and the second eigenmode function component is input into the bidirectional gated cyclic unit to extract temporal features. Then, the extracted spatial features and temporal features are input into the fully connected layer for feature reconstruction, the fault prediction results are output, and the model is trained according to the prediction results and the fault label cycle until the model converges. In this embodiment, variational mode decomposition and mutual information entropy analysis make the feature extraction process more efficient and accurate. At the same time, by using the bidirectional gated cyclic unit and the graph convolution network in parallel, the distribution network fault model can capture temporal and spatial features and reconstruct them at the same time, which improves the accuracy and efficiency of model fault detection at the same time.

[0119] In one embodiment, Figure 2 As shown, the present invention also proposes a distribution network fault prediction method. The steps provided by the distribution network fault prediction method in this embodiment include:

[0120] S201, obtaining current fault data of the distribution network.

[0121] The current distribution network fault data, including time series data such as current and voltage, and the topological structure data of the distribution network are collected as inputs to the model.

[0122] S202, applying variational mode decomposition to decompose the current fault data to obtain multiple eigenmode function components of the current fault data.

[0123] The variational mode decomposition technique is used to decompose the current fault data to obtain multiple eigenmode function components. This step helps to extract multiple intrinsic components in the data, so that the multi-level features of the data can be separated and analyzed.

[0124] S203, inputting multiple eigenmode function components of the current fault data into the distribution network fault model for prediction to obtain the current fault prediction result.

[0125] The distribution network fault model is trained by the above-mentioned distribution network fault model training method, and the model includes a parallel bidirectional gated recurrent unit and a graph convolutional network, as well as a fully connected layer connecting the two. In actual prediction, the model can effectively integrate temporal features and spatial features to optimize the accuracy and speed of fault detection.

[0126] To further illustrate the beneficial effects of the present invention, a specific simulation case is used to illustrate the present invention. Figure 3 The IEEE 33-node distribution network model is shown in the figure. There are 4 feeders and 33 nodes, and two wind distributed power modules are connected at the end of two of the feeders. Except for node 1, a fault point is set in the branches where the other 32 nodes are located. The fault type of each fault point is set to single-phase grounding (AG, BG, CG), two-phase short circuit (AB, AC, BC), two-phase short circuit grounding (ABG, ACG, BCG) and three-phase short circuit (ABC). The fault transition resistance is set to 0.01Ω, 5Ω, 50Ω, 500Ω, 1000Ω, 1500Ω, 2000Ω. Six initial fault phase angles are randomly generated each time, ranging from 1° to 180°. The final data set contains 13440 samples, 1 / 6 of which are randomly selected as the test set and the rest as the training set.

[0127] As shown in Table 1, Table 1 is a performance comparison table of the present invention and the control group model.

[0128]

[0129] Table 1

[0130] In Table 1, the graph convolutional network-bidirectional gated recurrent unit is our distribution network fault model. We can see that the parameter quantity and training time indicators of the algorithm model proposed in the present invention are significantly better than those of other control group models. In summary, the algorithm proposed in the present invention has good performance and is better than similar algorithms.

[0131] Figure 4 FIG. 2 shows an internal structure diagram of a terminal device in an embodiment. Figure 4As shown, the terminal device includes a processor, a memory and a network interface connected via a system bus. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the terminal device stores an operating system and may also store a computer program, which, when executed by the processor, enables the processor to implement a method. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement a method. Those skilled in the art will appreciate that Figure 4 The structure shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the terminal device to which the scheme of the present application is applied. The specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0132] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented: obtaining sample fault data of a distribution network; wherein the sample fault data carries a corresponding fault label; applying variational mode decomposition to decompose the sample fault data to obtain multiple intrinsic mode function components of the sample fault data; performing mutual information entropy analysis on the multiple intrinsic mode function components of the sample fault data to obtain an adjacency matrix and a feature matrix, and constructing a graph convolution network based on the adjacency matrix and the feature matrix; constructing a distribution network fault model; wherein the distribution network fault model includes a parallel bidirectional gated recurrent unit and a graph convolution network. A convolution network and a fully connected layer connecting a bidirectional gated recurrent unit and a graph convolution network; inputting the first intrinsic mode function component and the adjacency matrix into the graph convolution network for feature extraction to extract spatial features, and inputting the second intrinsic mode function component into the bidirectional gated recurrent unit for feature extraction to extract temporal features; wherein the first intrinsic mode function component and the second intrinsic mode function component are any one of multiple intrinsic mode function components of the sample fault data; inputting the spatial features and the temporal features into the fully connected layer for reconstruction to obtain the output fault prediction results, and cyclically training the distribution network fault model according to the fault prediction results and the fault labels until the distribution network fault model converges.

[0133] And implement the following steps: obtain current fault data of the distribution network; apply variational mode decomposition to decompose the current fault data to obtain multiple eigenmode function components of the current fault data; input the multiple eigenmode function components of the current fault data into the distribution network fault model for prediction to obtain the current fault prediction result; wherein the distribution network fault model is trained by the above-mentioned distribution network fault model training method.

[0134] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program: obtaining sample fault data of a distribution network; wherein the sample fault data carries a corresponding fault label; applying variational mode decomposition to decompose the sample fault data to obtain multiple intrinsic mode function components of the sample fault data; performing mutual information entropy analysis on the multiple intrinsic mode function components of the sample fault data to obtain an adjacency matrix and a feature matrix, and constructing a graph convolution network based on the adjacency matrix and the feature matrix; constructing a distribution network fault model; wherein the distribution network fault model comprises a parallel bidirectional gated loop A ring unit and a graph convolutional network, and a fully connected layer connecting the bidirectional gated recurrent unit and the graph convolutional network; inputting the first intrinsic mode function component and the adjacency matrix into the graph convolutional network for feature extraction to extract spatial features, and inputting the second intrinsic mode function component into the bidirectional gated recurrent unit for feature extraction to extract temporal features; wherein the first intrinsic mode function component and the second intrinsic mode function component are any one of multiple intrinsic mode function components of sample fault data; inputting the spatial features and the temporal features into the fully connected layer for reconstruction to obtain the output fault prediction results, and cyclically training the distribution network fault model according to the fault prediction results and the fault labels until the distribution network fault model converges.

[0135] And implement the following steps: obtain current fault data of the distribution network; apply variational mode decomposition to decompose the current fault data to obtain multiple eigenmode function components of the current fault data; input the multiple eigenmode function components of the current fault data into the distribution network fault model for prediction to obtain the current fault prediction result; wherein the distribution network fault model is trained by the above-mentioned distribution network fault model training method.

[0136] It should be noted that the above-mentioned distribution network fault prediction method, model training method, medium and equipment belong to a general inventive concept, and the contents in the distribution network fault prediction method, model training method, medium and equipment embodiments are applicable to each other.

[0137] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0138] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for a distribution network fault model, characterized in that: The method comprises: Acquire sample fault data of the distribution network; wherein the sample fault data carries a corresponding fault label; Decomposing the sample fault data by using variational mode decomposition to obtain a plurality of eigenmode function components of the sample fault data; Performing mutual information entropy analysis on multiple eigenmode function components of sample fault data to obtain an adjacency matrix and a feature matrix, and constructing a graph convolution network based on the adjacency matrix and the feature matrix; Constructing a distribution network fault model; wherein the distribution network fault model includes a parallel bidirectional gated recurrent unit and a graph convolutional network, and a fully connected layer connecting the bidirectional gated recurrent unit and the graph convolutional network; Inputting the first eigenmode function component and the adjacency matrix into a graph convolutional network for feature extraction to extract spatial features, and inputting the second eigenmode function component into a bidirectional gated recurrent unit for feature extraction to extract temporal features; wherein the first eigenmode function component and the second eigenmode function component are any one of a plurality of eigenmode function components of sample fault data; The spatial features and the temporal features are input into the fully connected layer for reconstruction, and the output fault prediction results are obtained. The distribution network fault model is cyclically trained according to the fault prediction results and the fault labels until the distribution network fault model converges.

2. The method according to claim 1, characterized in that: The applying variational mode decomposition to decompose the sample fault data to obtain a plurality of eigenmode function components of the sample fault data includes: For the sample fault data, different modes are set to have corresponding center frequencies and bandwidths, and a constraint model is set; wherein the constraint model is used to minimize the sum of the estimated bandwidths of each mode, and the constraint model is expressed as: In the above formula, K is the number of preset modes, {u k } is the kth mode component after decomposition, {ω k } is the kth center frequency after decomposition, δ(t) is the Dirac function; Based on the introduced penalty factor and Lagrange multiplication operator, the constrained model is converted into an unconstrained model, and the unconstrained model is expressed as: In the above formula, α is the penalty factor and λ is the Lagrange multiplication operator; Iteratively update the parameter u by alternating direction method of multipliers k ,ω k and λ, expressed as: In the above formula, and represents f(ω), u i (ω), λ(ω), and Fourier transform of , n is the number of iterations, γ is the noise tolerance; like Then return to execute the iterative update parameter u through the multiplier alternating direction method k ,ω k and λ steps and subsequent steps until To obtain K decomposed eigenmode function components.

3. The method according to claim 2, characterized in that The step of applying variational mode decomposition to decompose the sample fault data to obtain a plurality of eigenmode function components of the sample fault data further includes: Mutual information entropy is applied to perform correlation analysis on the intrinsic mode function components with the sample fault data as the target value to obtain multiple correlation values, which are expressed as: MI(IMF,X(t))=H(X(t))-H(X(t)|IMF) In the above formula, i = 1, 2, ..., K, mi represents the i-th correlation value, H(X(t)) represents the entropy of the sample fault data X(t), and H(X(t)|IMF) is the conditional entropy of the sample fault data X(t) when the intrinsic mode function component IMF is known; A preset number K′ of eigenmode function components with the highest correlation values ​​are retained; wherein K′<K.

4. The method according to claim 3, characterized in that The mutual information entropy analysis is performed on multiple eigenmode function components of the sample fault data to obtain an adjacency matrix and a characteristic matrix, including: Based on the mutual information entropy value between two nodes in each eigenmode function component, a K′×K′ connection matrix is ​​constructed as an adjacency matrix, where the mutual information entropy value is a measure of the degree of mutual dependence between two random variables; The absolute value of the normalized mutual information entropy between the eigenmode function components is used as the spatial feature of the adjacent components to construct a K′×F-dimensional feature matrix; where F is the length of the historical time series of the component.

5. The method according to claim 4, characterized in that The information transfer function of the graph convolutional network is expressed as: In the above formula, σ represents the Sigmoid activation function, H (l) represents the network output of layer l, W (l) represents the parameters to be trained, for The degree matrix, A represents the adjacency matrix, I N represents the N-order identity matrix, Represents the processed adjacency matrix.

6. The method according to claim 1, characterized in that The bidirectional gated recurrent unit model at time t is expressed as: r t =σ(W r ·[h t-1 ,x t ]) z t =σ(W z ·[h t-1 ,x t ]) In the above formula, r t is the reset gate, z t is the update gate, x t is the current input, is the current input and the previous hidden layer state summary, h t is the hidden layer output after updating the memory, W r is the weight matrix of the reset gate, W z is the weight matrix of the update gate, W h is the weight matrix of the hidden layer, σ is the Sigmoid function, and tanh is the hyperbolic tangent activation function.

7. The method according to claim 1, characterized in that The distribution network fault model also includes a channel attention mechanism layer, which is arranged between the bidirectional gated recurrent unit and the fully connected layer, and between the graph convolutional network and the fully connected layer. The method includes: applying the channel attention mechanism layer to remove the redundant information contained in the spatial features and the temporal features, which is specifically expressed as follows: Q=W Q *x i K=W K *x i V=W V *x i out i =sum(A i *V j ),j=1,2,...,n Among them, W Q Represents the weight matrix of the linear transformation of the query, W K Represents the weight matrix for linear transformation of keys, W V Represents the weight matrix for linear transformation of values, x i represents the i-th feature of the input, d K represents the dimension of the key vector, K T Indicates matrix transposition, out i Represents the i-th element of the output sequence, V j Represents the jth element of the value sequence, A i represents the attention weight of the i-th element.

8. A distribution network fault prediction method, characterized in that: The method comprises: Obtain current fault data of the distribution network; Decomposing the current fault data by applying variational mode decomposition to obtain a plurality of eigenmode function components of the current fault data; Input multiple eigenmode function components of current fault data into the distribution network fault model for prediction to obtain the current fault prediction result; wherein, the distribution network fault model is trained by any one of the methods according to claims 1-7.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 8.

10. A terminal device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 8.

Citation Information

Cited By

  • Fault identification method and system after high-proportion distributed power supply is connected to power distribution network, computer equipment and storage medium

    CN119441896A