Power distribution network fault tracing method based on graph convolutional neural network

Abstract

The invention proposes a power distribution network fault tracing method based on a graph convolutional neural network, and relates to the technical field of electric power, and the method comprises the steps: S1, constructing a power distribution network graph model, simulating a fault condition during the access of a high-proportion distributed power supply through the power distribution network graph model, so as to obtain power flow simulation data, and carrying out the data preprocessing; s2, constructing a graph convolutional neural network model containing a high-proportion distributed power supply according to the preprocessed power flow simulation data; s3, introducing the residual error into the image convolutional neural network model to construct a residual error-containing image convolutional neural network model; s4, carrying out the fault tracing of the power distribution network based on the residual image-containing convolutional neural network model; according to the invention, the node dependence problem in a complex power grid topological structure can be solved, the fault repair efficiency is improved, and the reliability and safety of power supply are realized.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a distribution network fault tracing method based on a graph convolutional neural network. Background Art

[0002] Distribution network fault tracing is a key link to ensure the stability of the power system and the reliability of power supply. It is of great significance to quickly and accurately determine the fault point of the distribution network, remove the fault line to avoid major power outages caused by line faults, reduce power outage losses and ensure the stable operation of the distribution network. The large-scale access of distributed energy to the power grid has become an irreversible trend, but it has also caused many problems in the fault tracing of the distribution network. First, the dynamic characteristics of distributed power sources make the characteristics of fault currents complicated. Traditional fault location methods rely on stable fault currents. In distributed systems, current waveforms are affected by multiple factors, resulting in fault signal fluctuations. Secondly, the widespread access of distributed power sources leads to changes in power flow and load distribution, making the fault propagation path more diverse and increasing the difficulty of positioning. In addition, the complexity of intelligent control systems requires positioning methods to have higher real-time and accuracy, so new technologies and strategies are urgently needed to meet these challenges.

[0003] Fault location in distribution networks is a key step in ensuring the stability and reliability of power supply. However, due to its complex topological structure and high coupling of electrical quantities, traditional neural networks and deep learning methods have significant limitations in processing non-Euclidean spatial data. Existing methods are difficult to fully capture the spatial correlation between nodes, and lack accuracy and robustness when dealing with data imbalance and noise interference. With the development of smart grids, the demand for more accurate and efficient fault location methods is becoming more urgent. Summary of the invention

[0004] The purpose of the present invention is to provide a distribution network fault tracing method based on graph convolutional neural network, which can solve the node dependency problem in complex power grid topology, improve the efficiency of fault repair, and achieve the reliability and safety of power supply.

[0005] The technical solution of the present invention is:

[0006] In a first aspect, the present application provides a distribution network fault tracing method based on a graph convolutional neural network, which comprises the following steps:

[0007] S1. Construct a distribution network diagram model, simulate the fault conditions when a high proportion of distributed power sources are connected through the distribution network diagram model to obtain power flow simulation data, and perform data preprocessing;

[0008] S2. Construct a graph convolutional neural network model containing a high proportion of distributed power sources based on the preprocessed power flow simulation data;

[0009] S3, introducing the residual into the graph convolutional neural network model to construct a graph convolutional neural network model containing residual;

[0010] S4. Fault tracing of distribution network based on residual graph convolutional neural network model.

[0011] Further, step S1 includes:

[0012] S11. Build a node model with distributed power sources;

[0013] S12, batch setting fault nodes and fault types in the node model and performing simulation calculations to obtain power flow simulation data;

[0014] S13, preprocessing the power flow simulation data.

[0015] Furthermore, the above data preprocessing process includes data cleaning and data normalization;

[0016] Among them, data cleaning includes the detection and processing of missing values ​​and outliers;

[0017] The calculation process of data normalization includes:

[0018] Minimum and maximum normalization:

[0019]

[0020] Z-score normalization:

[0021]

[0022] In the formula, is the normalized i-th feature, x i is the original value of the data, max(x i ) and min(x i ) represent the maximum and minimum values ​​of the i-th feature, respectively. is the normalized eigenvalue, X is the original value of the data, μ represents the mean of the data set, σ 标 Represents the standard deviation of the data set.

[0023] Furthermore, in step S2, the calculation process of constructing a graph convolutional neural network model containing a high proportion of distributed power sources includes:

[0024] The distribution network node feature vector feature matrix and adjacency matrix are taken as input:

[0025] Input=(X,A)

[0026] Construct the adjacency matrix:

[0027]

[0028] Calculate the nonlinear function of each convolutional layer:

[0029] H (l+1) =f(H (l) ,A),l=0,1,...,L

[0030] Calculate the layer conduction function of each layer of the network:

[0031] f(H (l) ,A)=σ(AH (l) W (l) )

[0032] Compute the matrix that is the sum of the adjacency matrix and the identity matrix:

[0033]

[0034] Compute the symmetric normalized matrix:

[0035]

[0036] Compute the degree matrix:

[0037]

[0038] The propagation formula of the graph convolutional neural network layer is obtained:

[0039]

[0040] In the formula, X represents the feature matrix, A represents the adjacency matrix, i and j represent the i-th and j-th nodes respectively, E represents the edge set, and H (l) represents the node features of the lth layer, l = 1, 2, 3, 4 respectively represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network, f represents the nonlinear function, L represents the number of convolutional layers, σ represents the activation function, and W (l) represents the weight matrix of the lth layer, I represents the identity matrix, represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the symmetric normalized adjacency matrix, D represents the degree matrix of A, D(i, j) represents the sum of the nodes in the i row or the j column, express The degree matrix of Represents the inverse square root matrix.

[0041] Furthermore, in step S3, the calculation process of constructing the residual graph convolutional neural network model includes a forward propagation process and a back propagation process;

[0042] Among them, the forward propagation process includes:

[0043] Calculate the first graph convolution layer:

[0044]

[0045] Perform nonlinear transformation on the input through the ReLU layer:

[0046]

[0047] Calculate the second graph convolution layer:

[0048]

[0049] Calculate the third graph convolution layer:

[0050]

[0051] Calculate the last layer output:

[0052] Y=Softmax(H (3) W (4) )

[0053] Calculate the residual mathematical expression:

[0054]

[0055] Calculate the i-th node, the loss function is:

[0056]

[0057] To prevent overfitting, add L2 regularization term:

[0058]

[0059] The total loss function is the average of all node losses plus a regularization term:

[0060]

[0061] The back propagation process is to calculate the gradient based on the loss between the prediction result calculated by the forward propagation and the true label to update the weight. The calculation process includes:

[0062] Calculate the third level error term:

[0063]

[0064] Calculate the second-layer and first-layer error terms:

[0065]

[0066] The weight gradients of the output layer, the third layer, the second layer, and the first layer are calculated as follows:

[0067]

[0068] In the formula, H (l) represents the node features of the lth layer, represents the symmetric normalized adjacency matrix, X represents the feature matrix, W (l) represents the weight matrix of the lth layer, x represents the input, Y represents the probability of each node belonging to a faulty or non-faulty node in the distribution network, σ represents the activation function, express The degree matrix of

[0069] represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the inverse square root matrix, L i represents the loss function of the i-th node, Y ic Represents the one-hot encoding of the true label, Y ic =1 means node i belongs to category c, represents the probability predicted by the model, w c is the weight of category c, ε is the value to prevent the logarithm from taking zero, L reg represents the regularization term, |W| F is the Frobenius norm of the matrix W, F in is the dimension of the input feature, that is, the number of features of each node, F out is the dimension of the output feature, that is, the number of output features, λ is the regularization coefficient, L 总 is the total loss function, N is the number of nodes, δ (l) is the error term of the lth layer, δ′(Z (l) ) is the derivative of the ReLU activation function of the lth layer, It represents the weight gradient of the lth layer, l is the number of convolutional layers of the graph convolutional neural network, and l=1, 2, 3, 4 represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network respectively.

[0070] In a second aspect, the present application provides an electronic device, including:

[0071] A memory for storing one or more programs;

[0072] processor;

[0073] When the above one or more programs are executed by the above processor, a distribution network fault tracing method based on a graph convolutional neural network as described in any one of the first aspects above is implemented.

[0074] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a distribution network fault tracing method based on a graph convolutional neural network as described in any one of the first aspects above.

[0075] Compared with the prior art, the present invention has at least the following advantages or beneficial effects:

[0076] (1) The present invention provides a distribution network fault tracing method based on graph convolutional neural network, which improves the computational efficiency by constructing a graph neural network model with residual connection to process power flow data, and solves the gradient vanishing and gradient exploding problems by introducing residuals, thereby ensuring good information while training a deeper network;

[0077] (2) Based on the feature aggregation results of the graph neural network model, the present invention deeply integrates the topological structure of the distribution network with the fault characteristics, transforms the distribution network fault tracing problem into a classification problem, solves the node dependency problem in the complex power grid topology structure, improves the fault repair efficiency, effectively improves the accuracy and timeliness of fault tracing, and achieves the reliability and safety of power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0079] Figure 1 A step diagram of a distribution network fault tracing method based on a graph convolutional neural network according to the present invention;

[0080] Figure 2 It is a schematic diagram of the data structure conversion of the distribution network diagram;

[0081] Figure 3 Schematic diagram of the graph convolutional neural network structure;

[0082] Figure 4 Schematic diagram of the structure of the residual block;

[0083] Figure 5 Schematic diagram of the connection of residual blocks;

[0084] Figure 6 It is a schematic diagram of the structure of the convolutional neural network model containing residual graph;

[0085] Figure 7 This is a flow chart of distribution network fault tracing with residual graph convolutional neural network model;

[0086] Figure 8 The figure is a schematic structural block diagram of an electronic device according to an embodiment of the present invention.

[0087] Icon: 101, memory; 102, processor; 103, communication interface. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0089] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0090] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0091] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprises..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0092] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0093] Example 1

[0094] See also Figure 1 , Figure 1 Shown is a step diagram of a distribution network fault tracing method based on a graph convolutional neural network provided in an embodiment of the present application.

[0095] In the first aspect, the present application provides a distribution network fault tracing method based on graph convolutional neural network, such as Figure 1As shown, it includes the following steps:

[0096] S1. Construct a distribution network diagram model, simulate the fault conditions when a high proportion of distributed power sources are connected through the distribution network diagram model to obtain power flow simulation data, and perform data preprocessing;

[0097] S2. Construct a graph convolutional neural network model containing a high proportion of distributed power sources based on the preprocessed power flow simulation data;

[0098] S3, introducing the residual into the graph convolutional neural network model to construct a graph convolutional neural network model containing residual;

[0099] S4. Fault tracing of distribution network based on residual graph convolutional neural network model.

[0100] As a preferred implementation, step S1 includes:

[0101] S11. Build a node model with distributed power sources;

[0102] S12, batch setting fault nodes and fault types in the node model and performing simulation calculations to obtain power flow simulation data;

[0103] S13, preprocessing the power flow simulation data.

[0104] It should be noted that there is a natural correspondence between the topological structure of the power grid and the graph data structure. The complexity and diversity of the topological structure of the distribution network make the graph data structure an ideal choice for describing its state. Under the premise of ignoring the internal information of the components, the distribution network is simplified into a topological graph consisting of nodes and edges, such as Figure 2 , showing the process of converting electrical topology into a graph model suitable for graph convolutional neural network model recognition. In this graph model, nodes represent grid elements such as substations, load points, distributed generation resources, and edges represent overhead lines or cable lines connecting these elements. Each node contains the electrical characteristics corresponding to the element, such as the amplitude of three-phase voltage and current, the amplitude of three-phase current, active power, reactive power, and voltage phase angle. These characteristics constitute the GC. Through this graph model modeling method, it can effectively represent the basic properties of distribution network nodes and the complex interactions between nodes, providing a solid foundation for the simulation of power flow data. In addition, this graph structure provides the necessary framework for the implementation of the graph convolutional neural network model. The input feature matrix of the N model. According to the connection method of the graph data structure of the distribution network, the corresponding input adjacency matrix can be constructed.

[0105] As a preferred implementation, the process of data preprocessing includes data cleaning and data normalization;

[0106] It should be noted that the data generated by the simulation system often faces problems of missing and imbalance, so data preprocessing is indispensable. There is no standardized standard process for data preprocessing, and different algorithms and models need to adopt different processing strategies. When locating distribution network faults, abnormal data will directly affect the accuracy of fault location. The present invention will perform a preprocessing step on the simulation data and extract feature data suitable for distribution network fault location to ensure the effectiveness of model training. After the graph convolutional neural network model is established, these feature data will be directly input into the model for training, thereby improving the accuracy of fault location.

[0107] Among them, data cleaning includes the detection and processing of missing values ​​and outliers;

[0108] It should be noted that missing values ​​refer to the partial missing of feature data, which will affect the training of some models and lead to poor model results. General missing value processing includes: missing value discarding and missing value filling. The choice of which method depends on the nature of the data, the proportion of missing values, the distribution of the data, and the purpose of the analysis; outliers are generally manifested as data with values ​​that are too large or too small, and the reasons for their occurrence include: reduced reliability of components, failure of measuring instruments, and human operating errors.

[0109] During the data collection and organization process, due to operational errors, equipment problems or other situations, data may be missing or distorted, which in turn affects the training speed of the evaluation model. Consider normalizing the collected input feature data to eliminate the dimensional effects between different features. Common data normalization methods are as follows:

[0110] Minimum and maximum normalization:

[0111]

[0112] It should be noted that this method is very sensitive to outliers, and when new data is added, the maximum and minimum values ​​may change, which requires re-normalization, increasing the amount of calculation;

[0113] Z-score normalization:

[0114]

[0115] In the formula, is the normalized i-th feature, x i is the original value of the data, max(x i ) and min(x i ) represent the maximum and minimum values ​​of the i-th feature, respectively. is the normalized eigenvalue, X is the original value of the data, μ represents the mean of the data set, σ 标Represents the standard deviation of the data set.

[0116] It should be noted that this method can transform the data into a method that converts the mean of all eigenvalues ​​to 0 and the standard deviation to 1; normalization does not change the distribution shape of the data. If the original data is approximately normally distributed, the normalized data will have a mean of 0 and a standard deviation of 1, but still maintain its original distribution shape; and unlike minimum and maximum normalization, Z-score normalization is less affected by outliers, making it more robust in practical applications, so Z-score normalization is selected to normalize the input data.

[0117] As a preferred implementation, in step S2, the calculation process of constructing a graph convolutional neural network model containing a high proportion of distributed power sources includes:

[0118] The distribution network node feature vector feature matrix and adjacency matrix are taken as input:

[0119] Input=(X,A)

[0120] Construct the adjacency matrix:

[0121]

[0122] Calculate the nonlinear function of each convolutional layer:

[0123] H (l+1) =f(H (l) ,A),l=0,1,...,L

[0124] Calculate the layer conduction function of each layer of the network:

[0125] f(H (l) ,A)=σ(AH (l) W (l) )

[0126] Compute the matrix that is the sum of the adjacency matrix and the identity matrix:

[0127]

[0128] Compute the symmetric normalized matrix:

[0129]

[0130] Compute the degree matrix:

[0131]

[0132] The propagation formula of the graph convolutional neural network layer is obtained:

[0133]

[0134] In the formula, X represents the feature matrix, A represents the adjacency matrix, i and j represent the i-th and j-th nodes respectively, E represents the edge set, and H (l) represents the node features of the lth layer, l = 1, 2, 3, 4 respectively represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network, f represents the nonlinear function, L represents the number of convolutional layers, σ represents the activation function, and W (l) represents the weight matrix of the lth layer, I represents the identity matrix, represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the symmetric normalized adjacency matrix, D represents the degree matrix of A, D(i, j) represents the sum of the nodes in the i row or the j column, express The degree matrix of Represents the inverse square root matrix.

[0135] It should be noted that if Figure 3 The figure shows the general structure of the graph convolutional neural network model. The core of the graph neural network is to iteratively aggregate local feature information of the graph. The information contained in any node is closely related to the information of its adjacent nodes. By establishing an adjacency matrix, the role of adjacent nodes in fault diagnosis is considered and the features of adjacent nodes are extracted. From a structural point of view, the typical graph convolutional neural network structure contains several convolutional layers and finally outputs through a fully connected layer. Compared with ordinary neural networks, the main feature of graph convolutional neural networks is that their input data is graph data, which not only contains the feature information of the nodes themselves, but also contains an adjacency matrix representing the connection relationship of the graph structure.

[0136] As a preferred implementation, in step S3, the calculation process of constructing the residual graph convolutional neural network model includes a forward propagation process and a back propagation process;

[0137] Among them, the forward propagation process includes:

[0138] Calculate the first graph convolution layer:

[0139]

[0140] The input is transformed nonlinearly through the ReLU layer:

[0141]

[0142] Calculate the second graph convolution layer:

[0143]

[0144] Calculate the third graph convolution layer:

[0145]

[0146] Calculate the last layer output:

[0147] Y=Softmax(H (3) W (4) )

[0148] Calculate the residual mathematical expression:

[0149]

[0150] Calculate the i-th node, the loss function is:

[0151]

[0152] To prevent overfitting, add L2 regularization term:

[0153]

[0154] The total loss function is the average of all node losses plus a regularization term:

[0155]

[0156] The back propagation process is to calculate the gradient based on the loss between the prediction result calculated by the forward propagation and the true label to update the weight. The calculation process includes:

[0157] Calculate the third level error term:

[0158]

[0159] Calculate the second-layer and first-layer error terms:

[0160]

[0161] The weight gradients of the output layer, the third layer, the second layer, and the first layer are calculated as follows:

[0162]

[0163] In the formula, H (l) represents the node features of the lth layer, represents the symmetric normalized adjacency matrix, X represents the feature matrix, W (l) represents the weight matrix of the lth layer, x represents the input, Y represents the probability of each node belonging to a faulty or non-faulty node in the distribution network, σ represents the activation function, express The degree matrix of represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the inverse square root matrix, L i represents the loss function of the i-th node, Yic Represents the one-hot encoding of the true label, Y ic =1 means node i belongs to category c, represents the probability predicted by the model, w c is the weight of category c, ε is the value to prevent the logarithm from taking zero, L reg represents the regularization term, |W F is the Frobenius norm of the matrix W, F in is the dimension of the input feature, that is, the number of features of each node, F out is the dimension of the output feature, that is, the number of output features, λ is the regularization coefficient, L 总 is the total loss function, N is the number of nodes, δ (l) is the error term of the lth layer, δ′(Z (l) ) is the derivative of the ReLU activation function of the lth layer, It represents the weight gradient of the lth layer, l is the number of convolutional layers of the graph convolutional neural network, and l=1, 2, 3, 4 represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network respectively.

[0164] like Figure 4 The figure shows the structure of the residual block. Inspired by the residual connection, the graph convolutional network and residual learning are combined. Residual connections are added between several graph convolutional layers to form a residual block, and then multiple residual blocks are combined to form a residual graph convolutional network. In this way, high-level feature information of graph data can be learned while avoiding network degradation problems. The basic principle of the graph neural network with residual connections is to solve the gradient vanishing and gradient exploding problems in deep network training by introducing residual connections on graph structure data. In the graph neural network model with residual connections, residual connections allow node features to be directly propagated across several layers, thereby maintaining the flow of information and accelerating the back propagation of gradients. This structure is particularly suitable for graph data because graph data has a complex topological structure and the features of nodes are affected not only by direct neighbors but also by more distant neighbors.

[0165] like Figure 5 The figure shows the connection diagram of the residual block. Introducing residuals for convolution calculation in the graph neural network can enhance the information complementarity between the deep and shallow layers. In addition, the convolution network that introduces residuals requires an identity mapping mechanism, which ensures the compatibility of shallow and deep information, makes the network structure more uniform, and is conducive to improving learning accuracy by deepening the network layer.

[0166] It should be noted that the present invention uses the Adadelta algorithm to update the weights in order to effectively update the model parameters. The algorithm avoids the complexity of setting the learning rate in the traditional method by dynamically adjusting the learning rate, making the model training more efficient and stable; and the Adadelta algorithm is an effective adaptive learning rate optimization algorithm that can improve the convergence speed and performance of the model in deep learning.

[0167] In summary, the present invention provides a distribution network fault tracing method based on a graph convolutional neural network. By constructing a distribution network graph model, the distribution network graph model is used to simulate the fault situation when a high proportion of distributed power sources are connected to obtain power flow simulation data, and data preprocessing is performed. A graph convolutional neural network model containing a high proportion of distributed power sources is constructed according to the preprocessed power flow simulation data. The collected power flow simulation data is batch normalized by the graph convolutional neural network model, and the normalized data is input for feature extraction. After convolution operations of three convolutional layers, the graph convolutional neural network model effectively extracts the relationship features between nodes and the topological structure information of the network, so that the final feature vector contains more information, and then the residual is introduced to establish a residual graph convolutional neural network model, such as Figure 6 As shown in Figure 1, the residual block added between the first layer output and the third layer of the residual graph convolutional neural network model solves the gradient vanishing and gradient exploding problems in deep network training, improves the robustness of the model, and finally the fully connected layer uses the softmax activation function to process the output and convert the output into a classification result, i.e., the fault location label. The flow chart of the distribution network fault tracing with the residual graph convolutional neural network model is shown in Figure 1. Figure 7 shown.

[0168] Example 2

[0169] See also Figure 8 , Figure 8 A schematic structural block diagram of an electronic device provided in an embodiment of the present application.

[0170] An electronic device includes a memory 101, a processor 102 and a communication interface 103, wherein the memory 101, the processor 102 and the communication interface 103 are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used to communicate signaling or data with other node devices.

[0171] Among them, the memory 101 can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.

[0172] The processor 102 may be an integrated circuit chip with signal processing capability. The processor 102 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0173] It can be understood that the structure shown in the figure is only for illustration, and a distribution network fault tracing method based on graph convolutional neural network may also include more or fewer components than shown in the figure, or have a different configuration than shown in the figure. Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0174] In the embodiments provided in the present application, it should be understood that the disclosed method can also be implemented in other ways. The embodiments described above are merely schematic, for example, the flowchart or block diagram in the accompanying drawings shows the possible implementation architecture, functions and operations of the method and computer program product according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of the boxes in the block diagram and / or the flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0175] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0176] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0177] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0178] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A distribution network fault tracing method based on graph convolutional neural network, characterized in that: The following steps are involved: S1. Construct a distribution network diagram model, simulate the fault conditions when a high proportion of distributed power sources are connected through the distribution network diagram model to obtain power flow simulation data, and perform data preprocessing; S2. Construct a graph convolutional neural network model containing a high proportion of distributed power sources based on the preprocessed power flow simulation data; S3, introducing the residual into the graph convolutional neural network model to construct a graph convolutional neural network model containing residual; S4. Fault tracing of distribution network based on residual graph convolutional neural network model.

2. A distribution network fault tracing method based on graph convolutional neural network as claimed in claim 1, characterized in that: Step S1 includes: S11. Build a node model with distributed power sources; S12, batch setting fault nodes and fault types in the node model and performing simulation calculations to obtain power flow simulation data; S13, preprocessing the power flow simulation data.

3. A distribution network fault tracing method based on graph convolutional neural network as claimed in claim 2, characterized in that: The data preprocessing process includes data cleaning and data normalization; Among them, data cleaning includes the detection and processing of missing values ​​and outliers; The calculation process of data normalization includes: Minimum and maximum normalization: Z-score normalization: In the formula, is the normalized i-th feature, x i is the original value of the data, max(x i ) and min(x i ) represent the maximum and minimum values ​​of the i-th feature, respectively. is the normalized eigenvalue, X is the original value of the data, μ represents the mean of the data set, σ 标 Represents the standard deviation of the data set.

4. A distribution network fault tracing method based on graph convolutional neural network as claimed in claim 1, characterized in that: In step S2, the calculation process of constructing a graph convolutional neural network model containing a high proportion of distributed power sources includes: The distribution network node feature vector feature matrix and adjacency matrix are taken as input: Input=(X,A) Construct the adjacency matrix: Calculate the nonlinear function of each convolutional layer: H (l+1) =f(H (l) ,A),l=0,1,...,L Calculate the layer conduction function of each layer of the network: f(H (l) ,A)=σ(AH (l) W (l) ) Compute the matrix that is the sum of the adjacency matrix and the identity matrix: Compute the symmetric normalized matrix: Compute the degree matrix: The propagation formula of the graph convolutional neural network layer is obtained: In the formula, X represents the feature matrix, A represents the adjacency matrix, i and j represent the i-th and j-th nodes respectively, E represents the edge set, and H (l) represents the node features of the lth layer, l = 1, 2, 3, 4 respectively represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network, f represents the nonlinear function, L represents the number of convolutional layers, σ represents the activation function, and W (l) represents the weight matrix of the lth layer, I represents the identity matrix, represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the symmetric normalized adjacency matrix, D represents the degree matrix of A, D(i, j) represents the sum of the nodes in the i row or the j column, express The degree matrix of Represents the inverse square root matrix.

5. A distribution network fault tracing method based on graph convolutional neural network as claimed in claim 4, characterized in that: In step S3, the calculation process of constructing the residual graph convolutional neural network model includes a forward propagation process and a back propagation process; Among them, the forward propagation process includes: Calculate the first graph convolution layer: The input is transformed nonlinearly through the ReLU layer: Calculate the second graph convolution layer: Calculate the third graph convolution layer: Calculate the last layer output: Y =Softmax(H (3) W (4) ) Calculate the residual mathematical expression: Calculate the i-th node, the loss function is: To prevent overfitting, add L2 regularization term: The total loss function is the average of all node losses plus a regularization term: The back propagation process is to calculate the gradient based on the loss between the prediction result calculated by the forward propagation and the true label to update the weight. The calculation process includes: Calculate the third level error term: Calculate the second-layer and first-layer error terms: The weight gradients of the output layer, the third layer, the second layer, and the first layer are calculated as follows: In the formula, H (l) represents the node features of the lth layer, represents the symmetric normalized adjacency matrix, X represents the feature matrix, W (l) represents the weight matrix of the lth layer, x represents the input, Y represents the probability of each node belonging to a faulty or non-faulty node in the distribution network, σ represents the activation function, express The degree matrix of represents the matrix resulting from the addition of the adjacency matrix and the identity matrix, represents the inverse square root matrix, L i represents the loss function of the i-th node, Y ic Represents the one-hot encoding of the true label, Y ic =1 means node i belongs to category c, represents the probability predicted by the model, w c is the weight of category c, ε is the value to prevent the logarithm from taking zero, L reg represents the regularization term, |W| F is the Frobenius norm of the matrix W, F in is the dimension of the input feature, that is, the number of features of each node, F out is the dimension of the output feature, that is, the number of output features, λ is the regularization coefficient, L 总 is the total loss function, N is the number of nodes, δ (l) is the error term of the lth layer, δ′(Z (l) ) is the derivative of the ReLU activation function of the lth layer, It represents the weight gradient of the lth layer, l is the number of convolutional layers of the graph convolutional neural network, and l=1, 2, 3, 4 represent the 1st, 2nd, 3rd, and output layers of the graph convolutional neural network respectively.

6. An electronic device, characterized in that: include: A memory for storing one or more programs; processor; When the one or more programs are executed by the processor, a distribution network fault tracing method based on a graph convolutional neural network as described in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, a distribution network fault tracing method based on a graph convolutional neural network as described in any one of claims 1 to 5 is implemented.

Images