Electronic load circuit abnormality diagnosis method based on dual-channel graph convolutional network

Abstract

The present application proposes an electronic load circuit anomaly diagnosis method based on a dual-channel graph convolutional network. Each component in the electronic load circuit is taken as a graph node, and the actual connection relationship of the components is taken as an edge. The electronic load circuit is converted into a graph structure, and the eigenvector of the graph node is the frequency domain information of the current and voltage of the device; a degree matrix is ​​created based on the graph structure, and the diagonal elements of the degree matrix are determined by the number of edges connecting each node. An adjacency matrix is ​​created based on the graph structure, and its value is determined by the connection relationship between the nodes; a graph convolutional neural network GCN is used to process the current and voltage data separately in two channels; a channel attention mechanism is used to fuse the feature data obtained from the two channels; and an SVM large-margin classification model is used for anomaly diagnosis. The present application proposes an electronic load circuit anomaly diagnosis method based on a dual-channel graph convolutional network, which realizes the anomaly detection of each component of the electronic load circuit.

Description

Technical Field

[0001] This application belongs to the field of machine learning and electronic loads, and specifically designs an electronic load circuit abnormality diagnosis method based on a dual-channel graph convolutional network. Background Art

[0002] Electronic load circuits are widely used in power electronics and battery testing, but long-term operation can lead to anomalies such as current and voltage fluctuations or component damage. Prompt diagnosis of these anomalies is crucial to improve stability and reliability.

[0003] Traditional diagnostic methods rely on expert experience, threshold settings, or signal processing techniques, but their effectiveness is limited in complex environments. Machine learning methods such as support vector machines (SVMs) and convolutional neural networks (CNNs), while offering certain advantages, struggle to capture topological relationships between components. Graph neural networks (GCNs) excel at processing topological data, improving anomaly detection accuracy. This paper proposes a diagnostic method based on a dual-channel graph convolutional network (GCN). This method represents circuit states through a graph structure and utilizes a dual-channel mechanism to extract current and voltage information separately, enhancing anomaly detection capabilities. Summary of the Invention

[0004] The purpose of this application is to provide an intelligent method to detect abnormalities in various components of an electronic DC load circuit and ensure timely detection of electronic load failures.

[0005] To solve the above technical problems, this application proposes an electronic load circuit abnormality diagnosis method based on a dual-channel graph convolutional network, including:

[0006] A circuit connection information graph is established based on the connection relationship of circuit components, with each node carrying a feature vector;

[0007] Obtain a degree matrix and an adjacency matrix based on the circuit connection information graph, and establish a current and voltage feature matrix based on the current and voltage feature vectors corresponding to each node in the circuit connection information graph;

[0008] A graph convolutional neural network is used to convolve the degree matrix, adjacency matrix, and feature matrix into two channels: voltage and current, to obtain two deep feature matrices.

[0009] The channel attention mechanism is used to fuse the two deep feature matrices of current and voltage;

[0010] The SVM large-margin classification model is used to perform abnormal diagnosis of electronic load circuits.

[0011] Optionally, the circuit connection information graph is established according to the connection relationship of the circuit components, and each node carries two feature vectors, including:

[0012] The current and voltage information of each component is sampled, and the current and voltage information are transformed into frequency domain information through spectrum conversion. Each device in the electronic DC load is defined as a node carrying current and voltage frequency domain information, and the connection relationship between devices is defined as an edge. The current and voltage frequency domain information of the device is used as the two eigenvectors of the node. The nodes carrying the eigenvectors are connected based on the edge relationship to form an electronic load circuit connection information graph.

[0013] Optionally, obtaining a degree matrix and an adjacency matrix according to the circuit connection information graph, and establishing a current and voltage feature matrix based on the current and voltage feature vectors corresponding to each node in the circuit connection information graph, include:

[0014] The degree matrix is ​​composed of a diagonal matrix converted from a sequence consisting of the number of edges connected to each node plus 1. The adjacency matrix is ​​composed of a "0\1" matrix representing the connection relationship plus a unit matrix of the same dimension. Several frequency values ​​are taken with equal steps in the frequency domain, and the current and voltage data of each component corresponding to these frequency values ​​are used to form the current and voltage characteristic matrices corresponding to the node.

[0015] Optionally, the graph convolutional neural network is used to convolve the degree matrix, adjacency matrix and feature matrix in two channels, voltage and current, to obtain two deep feature matrices, including:

[0016] Two parallel two-layer graph convolutional networks (GCNs) are built. The degree matrix, adjacency matrix, and current feature matrix are input into the current channel for feature aggregation calculation to obtain the current deep feature matrix. The degree matrix, adjacency matrix, and voltage feature matrix are input into the voltage channel for feature aggregation calculation to obtain the voltage deep feature matrix.

[0017] Optionally, the channel attention mechanism is used to fuse the two deep feature matrices of current and voltage, including:

[0018] The current and voltage deep feature matrices are subjected to global average pooling (GAP) to obtain two average eigenvalues. These two values ​​form a column vector. This column vector is passed through a two-layer fully connected network to obtain the attention weights corresponding to the current and voltage channels. Finally, the two deep feature matrices are weightedly fused to obtain the total feature matrix.

[0019] Optionally, the use of the SVM large-interval classification model to perform abnormality diagnosis on the electronic load circuit includes:

[0020] Based on the training data, we trained ten decision functions corresponding to the hyperplanes that can separate normal and abnormal data points and have the largest distance to normal and abnormal data points. Each row of the total feature matrix is ​​input into the corresponding decision function to obtain ten decision values. Then, we use the Platt scaling method to predict the abnormality probability.

[0021] The present application proposes an electronic load circuit anomaly diagnosis method based on a dual-channel graph convolutional network. Each component in the electronic load circuit is taken as a graph node, and the actual connection relationship of the components is taken as an edge. The electronic load circuit is converted into a graph structure, and the eigenvector of the graph node is the frequency domain information of the current and voltage of the device; a degree matrix is ​​created based on the graph structure, and the diagonal elements of the degree matrix are determined by the number of edges connecting each node. An adjacency matrix is ​​created based on the graph structure, and its value is determined by the connection relationship between the nodes; a graph convolutional neural network GCN is used to process the current and voltage data separately in two channels; a channel attention mechanism is used to fuse the feature data obtained from the two channels; and an SVM large-margin classification model is used for anomaly diagnosis. The present application proposes an electronic load circuit anomaly diagnosis method based on a dual-channel graph convolutional network, which realizes the anomaly detection of each component of the electronic load circuit. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0023] Figure 1 A schematic diagram of a process for detecting abnormalities in electronic load circuits based on a dual-channel graph convolutional network, which is provided in an embodiment of the present application;

[0024] Figure 2 An electronic load circuit provided by an embodiment of the present invention;

[0025] Figure 3 A circuit connection information diagram converted from the circuit diagram provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0027] Figure 1 The electronic load circuit abnormality diagnosis method based on a dual-channel graph convolutional network provided in the embodiment of the present application implements a flow chart of abnormality detection of various components of the electronic load circuit, which specifically includes five contents.

[0028] S11: Establish a circuit connection information graph based on the connection relationship of circuit components, with each node carrying two feature vectors.

[0029] It should be noted that the components in the electronic load circuit have an actual physical connection relationship. Based on this connection relationship, the working states of the components affect each other. If a part fails, its connection part will inevitably be disturbed. This application chooses to convert the actual circuit topology into a circuit connection information diagram for processing.

[0030] Step 11: Figure 2 The current and voltage information of the ten components R1, R2, R3, R4, programmable DA, MCU, BG1, BG2, BG3 and sampling resistor in the electronic load circuit at the same time segment are converted into frequency domain information using the fast Fourier transform shown in formula (1);

[0031]

[0032] Step 12: Figure 2 The ten components in the electronic load circuit, R1, R2, R3, R4, programmable DA, MCU, BG1, BG2, BG3 and sampling resistor, are defined as ten graph nodes;

[0033] Step 13: Take n points of the effective frequency range of the current and voltage frequency domain information of each component and average them, and use the frequency domain data value of the current and voltage corresponding to each point as the current and voltage characteristic vectors of the node corresponding to the component;

[0034] Step 14: Figure 2 The connection relationship between the components in the electronic load circuit diagram is defined as edges. These edges are used to connect the ten nodes carrying the eigenvectors, and the following is obtained: Figure 3 Circuit connection information diagram shown.

[0035] Based on the above discussion, in an optional embodiment of the present application, the number of graph nodes is 10, and the dimension n of the current and voltage feature vectors corresponding to each node is 1000.

[0036] S12: Obtain a degree matrix and an adjacency matrix according to the circuit connection information graph, and establish a current and voltage feature matrix based on the current and voltage feature vectors corresponding to each node in the circuit connection information graph.

[0037] It should be noted that the data in the circuit connection information diagram needs to be represented in a digital form before calculations can be performed. This application chooses to obtain the degree matrix and adjacency matrix representing its characteristics based on the circuit connection information diagram, and establish the current and voltage characteristic matrices based on the characteristic vectors obtained from the collected data.

[0038] Step 21: Figure 3 The ten nodes in the circuit connection information graph are numbered 0 to 9. From the graph, we can get the number of edges connected to each of the ten nodes. Arrange them in the order of 0 to 9 into a sequence and add 1 to each item.

[0039] Step 22: Convert this sequence into a diagonal matrix. This diagonal matrix is ​​the degree matrix of the circuit connection information graph. ;

[0040] Step 23: According to Figure 3 The circuit connection information diagram and the number of each node are used to represent the connection relationship of each node, and the number of pairs is equal to the number of edges;

[0041] Step 24: Construct a 10*10 zero matrix, whose rows and columns represent the ten nodes from 0 to 9 respectively. Assign the value of 1 to the position corresponding to the constructed number pairs in the matrix. Swap the two numbers of all the number pairs to obtain a new set of number pairs. Perform the assignment operation as above. Add a unit matrix of the same dimension to the matrix obtained by the two assignments to obtain the final adjacency matrix. ;

[0042] Step 25: Concatenate the current characteristic vectors of the ten nodes 0 to 9 in order from top to bottom into a 10*n matrix to form the current characteristic matrix of the circuit connection information diagram , similarly, we get the voltage characteristic matrix ;

[0043] Based on the above discussion, in an optional embodiment of the present application, a total of four matrices are obtained, the matrix and They are all 10*10 square arrays, and It is a 10*1000 matrix.

[0044] S13: Use graph convolutional neural network to convolve the degree matrix, adjacency matrix and feature matrix in two channels of voltage and current to obtain two deep feature matrices.

[0045] It should be noted that the working status of components has different effects on their current and voltage signals. This application chooses to process the current and voltage information in two channels.

[0046] Step 31: According to the degree matrix and the adjacency matrix , calculated as shown in formula (2) ;

[0047]

[0048] Step 32: Convert the current characteristic matrix and voltage characteristic matrix Input two GCN channels for calculation respectively, as shown in formulas (3)(4) and (5)(6), and obtain two deep feature matrices of current and voltage Record it as ;

[0049]

[0050]

[0051] Based on the above discussion, in an optional embodiment of the present application, the parameter matrix The dimension is 1000*700, the parameter matrix The dimension is 700*300, and the final deep feature matrix Both are 10*300 matrices, is the activation function.

[0052] S14: Use the channel attention mechanism to fuse the two deep feature matrices of current and voltage.

[0053] It should be noted that the deep feature matrices obtained after processing the current and voltage information have different contributions to the final abnormality diagnosis, so a set of weights is needed to process and fuse the two. This application selects the channel attention mechanism to fuse the current and voltage deep feature matrices.

[0054] Step 41: As shown in formulas (7) and (8), perform global average pooling (GAP) on the current and voltage depth feature matrices. They are the two dimensions of the depth feature matrix, and two values ​​are obtained and combined into a two-dimensional column vector , as shown in formula (9);

[0055]

[0056]

[0057]

[0058] Step 42: As shown in formulas (10) and (11), The vector is input into a two-layer fully connected network for calculation to obtain a two-dimensional column vector , represents the weight vector of the attention weights of the current and voltage channels, as shown in formula (12);

[0059]

[0060]

[0061]

[0062] Step 43: As shown in formula (13), according to the channel attention weight vector The two weight values ​​in are used to fuse the current and voltage deep feature matrices to obtain the total feature matrix;

[0063]

[0064] Based on the above discussion, in an optional embodiment of the present application, 10 and 300 respectively, the learnable weights of the two-layer fully connected network The dimension is 1*10, the weight The dimension is 10*1, is the activation function.

[0065] S15: Use the SVM large-margin classification model to perform abnormal diagnosis on the electronic load circuit.

[0066] It should be noted that the anomaly detection of various components in the electronic load circuit is essentially close to a binary classification problem, and classification is performed based on the total feature matrix. This application selects the SVM large-margin classification model to calculate the classification probability.

[0067] Step 51: For the nonlinear binary classification problem of anomaly detection, the RBF kernel function is introduced as shown in formula (14);

[0068]

[0069] Step 52: Split the ten rows of data in the total feature matrix and use them to train ten groups of decision function parameters for classifying ten components. The parameters include the Lagrange multiplier of the support vector. , labels of support vectors , support vector , bias term , we get ten different parameters, which are the decision functions in formula (15);

[0070]

[0071] Step 53: Based on the decision function The obtained value is used to estimate the abnormal probability using the Platt Scaling method shown in formula (16). It is a parameter fitted from the training set by the maximum likelihood method. This method outputs a probability between 0 and 1 for each decision function value, indicating the abnormal probability of the corresponding component.

[0072]

[0073] Based on the above discussion, in an optional example of the present application, the dimension of the support vector is the same as the total feature matrix dimension of 300, and the number of other parameters in each group is also 300;

[0074] Specific examples are used in this application to illustrate the principles and implementation methods of the present invention. The description of the above examples is only intended to help explain the method and core concept of the present invention. It should be noted that for ordinary people in this technical field, various improvements and modifications can be made to the present invention without departing from the principles of the present invention, and such improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. An electronic load circuit anomaly diagnosis method based on a dual-channel graph convolutional network, characterized by: The current and voltage information of each component in the circuit at the same time segment is transformed using fast Fourier transform Convert to frequency domain information; Take n points on average in the effective frequency range of the current and voltage frequency domain information of each component, and store the frequency domain data values ​​of the current and voltage corresponding to each point as two row vectors as the current and voltage feature vectors of the node corresponding to the component; Connect the component nodes according to the actual connection relationship in the circuit diagram to form a circuit connection information diagram; Obtain a degree matrix and an adjacency matrix based on the circuit connection information graph, and establish a current and voltage feature matrix based on the current and voltage feature vectors corresponding to each node in the circuit connection information graph; A' is calculated based on the adjacency matrix A and the degree matrix D. The current and voltage feature matrices corresponding to each node information are divided into two channels and input into two double-layer GCNs to calculate four deep feature matrices in is the learnable parameter matrix, and σ is the activation function: Using the channel attention mechanism, the weight matrix corresponding to current and voltage is first calculated, and then the two deep feature matrices of current and voltage are weighted fused; Anomaly detection is a nonlinear binary classification problem, and the RBF kernel function is introduced here: Each row of the total feature matrix is ​​used to train the decision function parameters corresponding to each component. The parameters include the Lagrange multiplier α of the support vector i , the labels of the support vector y i , support vector x i , bias term b, the decision function is: Based on the values ​​obtained from the decision function, the Platt scaling method is used to estimate the abnormality probability. This method outputs a probability between 0 and 1 for each decision function value, indicating the abnormality probability of the corresponding component: A and B are the parameters obtained by fitting the training set using the maximum likelihood method.

2. The electronic load circuit abnormality diagnosis method based on a dual-channel graph convolutional network according to claim 1, characterized in that: The degree matrix and adjacency matrix are obtained according to the circuit connection information graph. The current and voltage feature matrices are established based on the current and voltage feature vectors corresponding to each node in the circuit connection information graph, including: The degree matrix is ​​composed of a diagonal matrix converted from the sequence consisting of the number of edges connected to each node plus 1. The adjacency matrix is ​​composed of a "0\1" matrix representing the connection relationship plus a unit matrix of the same dimension. Several frequency values ​​are taken with equal steps in the frequency domain, and the current and voltage data of each component corresponding to these frequency values ​​are used to form the current and voltage characteristic matrices corresponding to the node.

3. The electronic load circuit abnormality diagnosis method based on a dual-channel graph convolutional network according to claim 1, characterized in that: The channel attention mechanism is used to fuse the two deep feature matrices of current and voltage, including: The current and voltage deep feature matrices are subjected to global average pooling (GAP) to obtain two average eigenvalues. These two values ​​form a column vector. This column vector is passed through a two-layer fully connected network to obtain the attention weights corresponding to the current and voltage channels. Finally, the two deep feature matrices are weightedly fused to obtain the total feature matrix.

Images