Transformer fault diagnosis method, device, system and storage medium
By processing transformer vibration signals through the Gram angle field and CBAM-CFBLS model, the problems of low efficiency and poor real-time performance of traditional power transformer fault diagnosis are solved, and real-time and accurate diagnosis of transformer faults is achieved, which is suitable for fault monitoring of power transformers.
Patent Information
- Application Number
- CN202411324417.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2025-09-19
- Estimated Expiration
- 2044-09-23
AI Technical Summary
Traditional power transformer fault diagnosis methods have low diagnostic efficiency and accuracy and are not suitable for real-time monitoring. The deep learning model training time is long and cannot meet the real-time monitoring requirements of power transformers.
The Gram angle field and CBAM-CFBLS model are used for transformer fault diagnosis. By acquiring the vibration signals of the transformer core and winding, a CBAM-CFBLS transformer fault diagnosis model is constructed. The Gram angle field is used to process the vibration signals to generate feature images. The multi-attention mechanism and width learning system are combined for feature extraction and model training.
It realizes real-time and accurate diagnosis of transformer faults. The model is small and highly adaptable, can maintain high sensitivity and accuracy under complex working conditions, is suitable for different fault types and environments, and has a wide range of applications.
Smart Images

Figure CN119274015B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of electronic digital data processing technology, and in particular relates to a transformer fault diagnosis method, device, system and storage medium. Background Art
[0002] Power transformers, as essential power transmission and transformation equipment, carry the crucial responsibility of converting electrical energy. Ensuring the proper functioning of transformers is crucial to the reliable operation of the entire power system. Over time, coupled with environmental impacts, transformers can develop potential mechanical failures, with the windings and core being the most prone to failure. If not addressed promptly, these failures can lead to power outages, fires, and other accidents.
[0003] Traditional power transformer fault diagnosis typically relies on manual experience and fixed reference rules, resulting in low diagnostic efficiency, low accuracy, and significant waste of human and material resources. However, the rapid development of deep learning technology has led to its widespread application in power transformer fault diagnosis, effectively addressing these issues. However, deep neural networks require a deep network structure to ensure detection accuracy, resulting in large network models and inconvenient deployment. Furthermore, as data continues to grow, deep learning models require retraining, resulting in lengthy network detection times and failing to meet the requirements for real-time power transformer monitoring.
[0004] Therefore, there is an urgent need for a method that can diagnose transformer faults in real time and accurately. Summary of the Invention
[0005] The purpose of this application is to provide a transformer fault diagnosis method, device, system and storage medium, which realize transformer fault diagnosis based on Gram angle field and CBAM-CFBLS model, with the advantages of small model, high real-time performance and high diagnostic accuracy.
[0006] In order to achieve the above objectives, the solution of this application is:
[0007] In a first aspect, an embodiment of the present application provides a transformer fault diagnosis method, comprising:
[0008] Obtain vibration signals of transformer core and winding;
[0009] The vibration signal is processed by the Gram angle field to obtain a characteristic image, wherein the characteristic image includes a Gram angle sum field characteristic image and a Gram angle difference field characteristic image;
[0010] The Gram angle sum field feature image and the Gram angle difference field feature image are processed through a multi-attention mechanism to obtain a global feature matrix.
[0011] Based on the width learning system, the feature nodes of the global feature matrix are cascaded to build the CFBLS model;
[0012] A dataset was constructed based on the feature images and divided into a training set and a test set. The global feature matrix was input into the CFBLS model, and the parameters of the CFBLS model were adjusted using the training set to construct a CBAM-CFBLS transformer fault diagnosis model. The CBAM-CFBLS transformer fault diagnosis model was then tested using the test set until the fault diagnosis accuracy reached a preset threshold.
[0013] Transformer fault diagnosis is performed using the trained CBAM-CFBLS transformer fault diagnosis model.
[0014] The above method according to the embodiment of the present application may also have the following additional technical features:
[0015] Furthermore, the vibration signal is processed by the Gram angle field to obtain a feature image, including:
[0016] By the sequence X=(X1,X2,...,X n ) represents the vibration signal, normalizes the vibration signal, normalizes the sequence X to the interval [0,1], and obtains the normalized value The formula is as follows:
[0017]
[0018] Where i = 1, 2, ..., n, x i represents the i-th group of vibration signals, max(X) represents the minimum value in the vibration signal, and min(X) represents the minimum value in the vibration signal.
[0019] Furthermore, the method includes mapping the sequence X into a time series in polar coordinates, as shown below:
[0020]
[0021] Among them, t i represents the time node, N represents the regularization constant factor in polar coordinates, θ i represents the level angle of point i, r represents the polar coordinate radius, Represents the normalized time series.
[0022] Furthermore, the method includes removing noise from the vibration signal through two inner products with penalty terms, as shown below:
[0023]
[0024]
[0025] Here, <·> represents the inner product operation.
[0026] Furthermore, the method includes generating a Gram angle and field two-dimensional feature image by calculating the cosine function of the sum of each two level angles and summing them up, and the formula is as follows:
[0027]
[0028] By calculating the sine function of the difference between every two level angles and summing them up, a two-dimensional characteristic image of the Gram angle difference field is generated. The formula is as follows:
[0029]
[0030] Where I represents the unit row vector, represents the normalized time series, express The transpose of .
[0031] Furthermore, the Gram angle sum field feature image and the Gram angle difference field feature image are processed through a multi-attention mechanism to obtain a global feature matrix, including:
[0032] The Gram angle sum field feature image and the Gram angle difference field feature image are used as the input of the multi-attention mechanism. The multi-attention mechanism includes the channel attention mechanism and the spatial attention mechanism. The formula of the channel attention mechanism is as follows:
[0033]
[0034] Where i = 1, 2, ..., n, Z i ' represents the features of the i-th group of feature maps output by the channel attention mechanism, M c represents the attention weight in the channel dimension, It represents the feature map obtained after GAF transformation of the input i-th group of vibration signal samples. represents element-wise multiplication;
[0035] The features output by the channel attention mechanism are used as the input of the spatial attention mechanism. The formula is as follows:
[0036]
[0037] Among them, M s represents the attention weight in the spatial dimension; Z i " represents the features output by the spatial attention mechanism of the i-th group of feature maps;
[0038] The formula of the global feature matrix is as follows:
[0039]
[0040] Furthermore, based on the width learning system, the feature nodes of the global feature matrix are cascaded to construct the CFBLS model, including:
[0041] The global feature matrix is input into the width learning system. The width learning system converts the global feature matrix into n groups of random feature nodes through linear mapping, and uses the output of the previous group of feature nodes as the input of the next group of feature nodes to obtain the i-th group of feature nodes of the CFBLS model. The formula is as follows:
[0042]
[0043] Where i = 1, 2, ..., n, represents the feature layer activation function, W ei and β ei Respectively represent the weight and bias of the randomly generated i-th feature node;
[0044] Each set of feature nodes of the CFBLS model is connected in series to obtain a complete feature mapping layer Z p , the formula is as follows:
[0045] Z p =[Z1,Z2,...,Z p ]
[0046] For the complete feature map node Z p Perform nonlinear transformation to obtain enhanced nodes. The formula is as follows:
[0047] H j =ξ(ZW hk +β hk )
[0048] Where j = 1, 2, ..., q, ξ(·) represents the linear activation function of the enhancement node, W hk and β hk Represent the weight and bias of the randomly generated k-th enhancement node respectively;
[0049] Each enhancement node is connected in series to obtain the complete enhancement layer H q , the formula is as follows:
[0050] H q =[H1H2,...,H q ]
[0051] The feature map layer Z p and enhancement layer H q After being connected in series to the output layer, a complete CFBLS model is formed. The formula of the output matrix is as follows:
[0052] Y=[Z p |H q ]W=AW
[0053] Where A=[Z p |H q ], represents the hidden layer of all feature mapping layers and enhancement layers, and W represents the weight connecting the hidden layer and the output layer;
[0054] The objective function for calculating the optimal output weight is as follows:
[0055]
[0056] Where λ represents the regularization parameter;
[0057] Calculate the optimal output weight W by the Tychonov regularization method best , the formula is as follows:
[0058] W best =A + Y=(λI+A T A) -1 A T Y
[0059] Among them, A + represents the pseudo-inverse matrix of A, and I represents the identity matrix;
[0060] When λ→0, the pseudo-inverse matrix A + The formula is as follows:
[0061]
[0062] In a second aspect, an embodiment of the present application provides a transformer fault diagnosis device, comprising:
[0063] A signal acquisition module is configured to acquire vibration signals of the transformer core and winding;
[0064] A feature image module is configured to process the vibration signal through a Gram angle field to obtain a feature image, wherein the feature image includes a Gram angle sum field feature image and a Gram angle difference field feature image;
[0065] a feature matrix module configured to process the Gram angle sum field feature image and the Gram angle difference field feature image through a multi-attention mechanism to obtain a global feature matrix;
[0066] A model building module is configured to perform feature node cascading on a global feature matrix based on a width learning system to build a CFBLS model;
[0067] The model training module is configured to construct a data set based on the feature images and divide the data set into a training set and a test set; input the global feature matrix into the CFBLS model, adjust the parameters of the CFBLS model using the training set, construct a CBAM-CFBLS transformer fault diagnosis model, and test the CBAM-CFBLS transformer fault diagnosis model using the test set until the fault diagnosis accuracy reaches a preset threshold;
[0068] The fault diagnosis module is configured to perform transformer fault diagnosis using the trained CBAM-CFBLS transformer fault diagnosis model.
[0069] In a third aspect, an embodiment of the present application provides a transformer fault diagnosis system, the system including a processor and a memory, the memory storing a computer program, and the computer program being loaded and executed by the processor to implement the transformer fault diagnosis method provided in the first aspect of the embodiment of the present application.
[0070] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the transformer fault diagnosis method as described in the first aspect of the embodiment of the present application.
[0071] The transformer fault diagnosis method provided by the embodiment of the present application has the following beneficial technical effects compared with the prior art:
[0072] The present invention utilizes Gram Angular Field (GAF) technology to transform raw vibration signals into two-dimensional feature images, significantly improving the efficiency and accuracy of signal processing. The GASF and GADF feature images not only intuitively demonstrate the dynamic changes in the signal but also remove redundant information and noise in the time domain through angular domain conversion, thereby extracting purer and more effective features.
[0073] The embodiment of the present application introduces a multi-attention mechanism to automatically identify and strengthen key feature areas and channels, significantly enhancing the feature representation capability, enabling the model to more accurately capture subtle changes in vibration signals that are closely related to faults, and maintain high sensitivity and accuracy even under complex and changeable working conditions.
[0074] The present embodiment constructs a CFBLS model by improving the connection method of BLS feature nodes, reducing the model size. Through linear mapping and cascading of random feature nodes, CFBLS can quickly construct feature mapping layers and enhancement layers with rich representation capabilities. Simultaneously, the application of ridge regression and Tikhonov regularization further ensures the stability and generalization of model training. This efficient model training and optimization process enables the method to quickly adapt to different datasets and diagnostic tasks.
[0075] The embodiments of this application combine multiple advanced technologies to form a comprehensive and powerful transformer fault diagnosis solution. Its generalization capability is reflected not only in its accurate identification of different fault types, but also in its adaptability to different operating conditions and complex environments. Whether it is for old equipment or new transformers, this method can provide reliable and accurate fault diagnosis results. In addition, this method has a wide range of applications and can be extended to fault diagnosis of other types of power equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] Figure 1 A schematic diagram showing a flow chart of a transformer fault diagnosis method according to an embodiment of the present application is shown;
[0077] Figure 2 The following is a structural block diagram of a transformer fault diagnosis device according to an embodiment of the present application;
[0078] Figure 3 A structural block diagram of a computer device according to an embodiment of the present application is shown;
[0079] Figure 4 The transformer normal signal GAF conversion process of the embodiment of the present application is shown;
[0080] Figure 5 The core loosening vibration signal GAF conversion process of the embodiment of the present application is given;
[0081] Figure 6 The winding stacking vibration signal GAF conversion process of the embodiment of the present application is given;
[0082] Figure 7 The winding deformation vibration signal GAF conversion process of the embodiment of the present application is given;
[0083] Figure 8 The winding loose vibration signal GAF conversion process of the embodiment of the present application is given;
[0084] Figure 9 The vibration signal GAF conversion process of the embodiment of the present application in which the core loosening and the winding overlap occur simultaneously is shown;
[0085] Figure 10 The vibration signal GAF conversion process of the embodiment of the present application in which the core loosening and winding deformation occur simultaneously is shown;
[0086] Figure 11 The CBAM-CFBLS model structure of the embodiment of the present application is given;
[0087] Figure 12 The figure shows the diagnostic results of the embodiment of the present application when the test set is fed into the CBAM-CFBLS model. DETAILED DESCRIPTION
[0088] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some, rather than all, structures related to the present application are shown in the accompanying drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0089] As used herein, the terms "comprise," "comprising," and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.
[0090] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments.
[0091] The Gramian Angular Field (GAF) is an effective method for converting one-dimensional time series data into a two-dimensional image representation. By treating each data point in the one-dimensional time series as a point in vector space and calculating the cosine of the angles between these points, the GAF then maps these cosine values onto the pixels of a two-dimensional image, generating an image that reflects the dynamic and cyclical characteristics of the time series. This method leverages the concept of the Gramian matrix in linear algebra and measures the similarity of data at different times by calculating the inner product between vectors.
[0092] The Gramian Angular Summation Field (GASF) is a technique for converting one-dimensional time series data into two-dimensional images. It is an implementation of the Gramian Angular Field (GAF). GASF maps the one-dimensional time series data onto the unit circle and calculates the cosine sum of the angles between these points to generate a two-dimensional image that reflects the dynamic and cyclical characteristics of the time series. This method preserves the temporal dependencies and correlations in the time series, making the converted image suitable for further analysis and processing.
[0093] The Gramian Angular Difference Field (GADF) is a technique for converting one-dimensional time series data into a two-dimensional image representation. Based on the concept of the Gramian matrix, the GADF method calculates the angular differences between different moments in the time series data to generate a two-dimensional matrix that reflects the data's changing trends. This matrix is then used as the pixel values to generate a two-dimensional image. This method leverages the temporal nature of time series data and the geometric significance of angular differences to convert one-dimensional data into a two-dimensional image for further analysis and processing.
[0094] The Broad Learning System (BLS) is a neural network learning framework designed to improve learning ability and efficiency by expanding the width of the network rather than its depth. It is an improvement and development of the Random Vector Functional Linker Network (RVFLNN) and the Single-Layer Feedforward Neural Network (SLFN). BLS emphasizes feature extraction and linear transformation. Its basic structure consists of an input layer, a mapping node (feature node) layer, an enhancement node layer, and an output layer. Unlike deep learning, BLS does not increase the number of network layers, but instead increases the number of nodes in the same layer to achieve efficient incremental learning.
[0095] The Convolutional Block Attention Module (CBAM), a multi-attention mechanism, is an attention mechanism designed to enhance the modeling and representation capabilities of convolutional neural networks (CNNs) for image features. By introducing two mechanisms—channel attention and spatial attention—CBAM enables the model to dynamically adjust the weights of feature maps to adapt to different tasks and scenarios. CBAM is a lightweight attention module that can be seamlessly integrated into existing convolutional neural network architectures without requiring significant modifications to the underlying network structure. By sequentially applying channel attention and spatial attention, it enhances the network's focus on important features, thereby improving model performance.
[0096] The embodiment of the present application introduces a multi-attention mechanism (Convolutional Block Attention Module, CBAM), namely a channel attention mechanism and a spatial attention mechanism, before the feature node of the Broad Learning System (BLS), to obtain global features and improve the feature expression effect of BLS. By cascading feature nodes, CFBLS is constructed, the obtained global features are input into the CFBLS feature node, and the model is trained and adjusted with a training set to establish a CBAM-CFBLS (Convolutional Block Attention Module-enhanced Broad Learning System) transformer fault diagnosis model.
[0097] like Figure 1 As shown, the embodiment of the present application provides a transformer fault diagnosis method, comprising the following steps:
[0098] Step 101: Obtain vibration signals of the transformer core and windings.
[0099] During transformer operation, the core and windings vibrate due to various factors, including electromagnetic and mechanical forces. These vibration signals contain a wealth of information about the equipment's status. By collecting and analyzing these signals, transformer status monitoring and fault diagnosis can be achieved.
[0100] To comprehensively and accurately acquire vibration signals from the transformer core and windings, appropriate vibration measurement points must be selected on the transformer body surface. A vibration sensor must be installed at each selected vibration measurement point to collect the vibration signal. Vibration sensors typically include accelerometers and velocity sensors, and the appropriate sensor can be selected based on actual needs. Sensor installation must be stable and reliable to minimize the impact of external interference on the signal.
[0101] The present embodiment uses an acceleration sensor for on-site measurement to collect transformer vibration signals. Specifically, the transformer vibration signal fundamental frequency is 100 Hz, the transformer vibration signal sampling frequency is 10 kHz, and a set of vibration signal data is acquired every 10 seconds, that is, 1000 data points constitute a vibration signal sample. The collected transformer vibration signals include normal transformer vibration signals, core loosening vibration signals, winding overlap vibration signals, winding deformation vibration signals, winding loosening vibration signals, vibration signals of core loosening and winding overlap, and vibration signals of core loosening and winding deformation. A total of seven vibration signals, 300 vibration signal samples are collected for each vibration signal, for a total of 2100 vibration signal samples.
[0102] Step 102 : Process the vibration signal using the Gram angle field to obtain a feature image. The feature image includes a Gram angle sum field feature image and a Gram angle difference field feature image.
[0103] Vibration signal processing using the Gramian Angular Field (GAF) is a method for converting time series data into an image representation. This method is particularly suitable for processing one-dimensional time series data such as vibration signals. This method preserves the dynamic characteristics and correlations of the time series while converting it into a two-dimensional image, facilitating subsequent feature extraction and pattern recognition.
[0104] The embodiment of the present application uses GAF to convert the transformer vibration signal into a two-dimensional image, normalizes the collected transformer vibration signal, eliminates the noise in the transformer vibration signal through the inner product with a penalty term, and then completes the conversion of the transformer vibration signal into two feature images, GASF and GADF, by GAF.
[0105] Specifically, through the sequence X=(X1,X2,...,X n ) represents the vibration signal, normalizes the vibration signal, normalizes the sequence X to the interval [0,1], and obtains the normalized value The formula is as follows:
[0106] Where i = 1, 2, ..., n, x i represents the i-th group of vibration signals, max(X) represents the minimum value in the vibration signal, and min(X) represents the minimum value in the vibration signal.
[0107] The transformer vibration signal sequence X is mapped to a time series in polar coordinates by taking the time node as the arc cosine of the radius and scaling value to generate polar coordinates. The formula is as follows:
[0108]
[0109] Among them, t i represents the time node, N represents the regularization constant factor in polar coordinates, θ i represents the level angle of point i, r represents the polar coordinate radius, Represents the normalized time series.
[0110] The noise in the transformer vibration signal is eliminated by the inner product with penalty terms. The formulas for two unique inner products with penalty terms are as follows:
[0111]
[0112] Here, <·> represents the inner product operation.
[0113] The GASF two-dimensional characteristic graph is generated by calculating the cosine function of the sum of each two level angles and summing them up. The formula is as follows:
[0114]
[0115] By calculating the sine function of the difference between each two level angles and summing them up, a GADF two-dimensional feature map is generated. The formula is as follows:
[0116]
[0117] Where I represents the unit row vector, represents the normalized time series, express The above transformation converts the given time series into two-dimensional feature images of GASF and GADF that are symmetrical along the diagonal, with increased time and pixel points, preserving the time-related features within the time series.
[0118] The 2100 vibration signal samples were transformed through GAF to obtain 1400 GAF images, which included 7 categories and 300 vibration signal samples in each category.
[0119] Step 103: Process the Gram angle sum field feature image and the Gram angle difference field feature image through a multi-attention mechanism to obtain a global feature matrix.
[0120] When processing image features, especially when it comes to features such as Gramian Angular Field (GAF) and Gramian Angular Difference Field (GADF), using a multi-attention mechanism to extract a global feature matrix is an effective method. This method can help the model pay more attention to important areas in the image, thereby improving the quality of feature representation.
[0121] The embodiment of the present application introduces CBAM, namely the channel attention mechanism and the spatial attention mechanism, before the BLS feature node to obtain global features, improve the feature expression effect of BLS, and construct CFBLS by cascading feature nodes.
[0122] Specifically, the GASF and GADF feature images are used as CBAM inputs. First, the channel attention mechanism is used to suppress background noise and redundant information in the remote sensing image in the channel dimension, while enhancing the expression of significant features. The formula is as follows:
[0123]
[0124] Where i = 1, 2, ..., n, Z i ' represents the features of the i-th group of feature maps output by the channel attention mechanism, M c represents the attention weight in the channel dimension, It represents the feature map obtained after GAF transformation of the input i-th group of vibration signal samples. represents element-wise multiplication;
[0125] The features output by the channel attention mechanism are used as the input of the spatial attention mechanism. The formula is as follows:
[0126]
[0127] Among them, M s represents the attention weight in the spatial dimension; Z i " represents the features output by the spatial attention mechanism of the i-th group of feature maps;
[0128] After the above CBAM processing, the global feature matrix is obtained:
[0129] Step 104 : Based on the width learning system, feature nodes of the global feature matrix are cascaded to construct a CFBLS model.
[0130] Based on the Broad Learning System (BLS), the global feature matrix is cascaded with feature nodes to construct a Composite Feature Broad Learning System (CFBLS) model, which is a task that extends the traditional BLS framework to handle more complex feature fusion and data representation. The Broad Learning System is a fast learning framework based on a random weight neural network, which enhances the feature representation of the input data by constructing enhancement nodes. BLS mainly consists of three parts: the input layer, the enhancement layer (also called the feature mapping layer), and the output layer. The global feature matrix is usually obtained from the original data through feature extraction methods (such as PCA, LDA, deep learning feature extraction, etc.). This matrix contains a global representation of the data sample, each row represents a sample, and each column represents a feature. In CFBLS, the embodiment of the present application regards the global feature matrix as part of the input layer, and performs cascading and enhancement of feature nodes on this basis.
[0131] The BLS of the present application converts the input Converting to n groups of random feature nodes is done by taking the output of the previous group of feature nodes as the input of the next group of feature nodes. The formulas for the first and second groups of feature nodes are as follows:
[0132]
[0133] Where, represents the feature layer activation function, W e1 and β e1 Represent the weight and bias of the randomly generated first feature node, W e2 and β e2 Represent the weight and bias of the randomly generated second feature node respectively.
[0134] By analogy, we construct CFBLS. The formula for the i-th group of feature nodes in CFBLS is as follows:
[0135]
[0136] Where i = 1, 2, ..., n, represents the feature layer activation function, W ei and β ei They represent the weight and bias of the randomly generated i-th feature node respectively.
[0137] Step 105: construct a data set based on the feature image and divide the data set into a training set and a test set; input the global feature matrix into the CFBLS model, adjust the parameters of the CFBLS model using the training set, construct a CBAM-CFBLS transformer fault diagnosis model, and test the CBAM-CFBLS transformer fault diagnosis model using the test set until the fault diagnosis accuracy reaches a preset threshold.
[0138] In the embodiment of the present application, based on the vibration signal samples obtained in step 102, 300 samples of each type are divided into a training set and a test set in a ratio of 7:3.
[0139] like Figure 11 As shown, in order to construct a transformer fault diagnosis model based on feature images and apply the combination of CBAM and CFBLS, the embodiment of the present application connects each group of feature nodes of the CFBLS model in series to obtain a complete feature mapping layer Z p , the formula is as follows:
[0140] Z p =[Z1,Z2,...,Z p ]
[0141] For the complete feature map node Z p Perform nonlinear transformation to obtain enhanced nodes. This step aims to increase the nonlinear ability and feature expression ability of the model. The formula is as follows:
[0142] H j =ξ(ZW hk +βhk )
[0143] Where j = 1, 2, ..., q, ξ(·) represents the linear activation function of the enhancement node, W hk and β hk Represent the weight and bias of the randomly generated k-th enhancement node respectively;
[0144] Each enhancement node is connected in series to obtain the complete enhancement layer H q , the formula is as follows:
[0145] H q =[H1H2,...,H q ]
[0146] The feature map layer Z p and enhancement layer H q After being connected in series to the output layer, a complete CFBLS model is formed. The formula of the output matrix is as follows:
[0147] Y=[Z p |H q ]W=AW
[0148] Where A=[Z p |H q ], represents the hidden layer of all feature mapping layers and enhancement layers, and W represents the weight connecting the hidden layer and the output layer;
[0149] In terms of model training and optimization, the embodiment of the present application optimizes and solves the objective function through the ridge regression method. Ridge regression is a technique used to solve the overfitting problem in linear regression. It limits the complexity of the model by adding a regularization term (usually the sum of the squares of the weights) to the loss function and calculates the objective function of the optimal output weight. The formula is as follows:
[0150]
[0151] Where λ represents the regularization parameter;
[0152] The optimal output weight is calculated by the Tikhonov regularization method. Tikhonov regularization (also known as Tikhonov regularization) is a form of ridge regression. It is mathematically equivalent to adding a small positive definite matrix to the coefficient matrix when solving a linear system of equations to improve the stability of the solution and obtain the optimal output weight W. best , the formula is as follows:
[0153] W best =A + Y=(λI+A T A) -1 A T Y
[0154] Among them, A + represents the pseudo-inverse matrix of A, and I represents the identity matrix;
[0155] When λ→0, the pseudo-inverse matrix A + The formula is as follows:
[0156]
[0157] At this point, the embodiment of the present application has obtained the CBAM-CFBLS transformer fault diagnosis model through the above steps. Next, the transformer fault diagnosis model CBAM-CFBLS is tested using a test set to calculate the diagnostic accuracy of the model for the test set.
[0158] The fault information is shown in the following table:
[0159] Table 1
[0160] Fault location Fault type Label normal / 1 iron core loose 2 Winding Overlap 3 Winding Deformation 4 Winding loose 5 Core, winding Loose, overlapping 6 Core, winding Looseness, deformation 7
[0161] like Figures 4 to 10 As shown in the figure, there are normal transformer vibration signals, core loose vibration signals, winding overlap vibration signals, winding deformation vibration signals, winding loose vibration signals, vibration signals of core looseness and winding overlap, and vibration signals of core looseness and winding deformation. The seven signals are converted into GASF and GADF two-dimensional feature images through GAF transformation.
[0162] like Figure 12 Figure 2 shows the test results of the transformer fault diagnosis model CASM-CFBLSN using the test set.
[0163] Step 106: Perform transformer fault diagnosis using the trained CBAM-CFBLS transformer fault diagnosis model.
[0164] Ensure that the transformer data required for fault diagnosis matches the data format and type used when training the model. Typically, this data will be transformer features represented in the form of images (such as Gram angle field, Gram angle difference field, etc.).
[0165] The input data is preprocessed with the same steps as the training data, including normalization and scaling, to ensure data consistency.
[0166] Load the trained CBAM-CFBLS model into the working environment and input the preprocessed transformer data into the CBAM-CFBLS model. Within the model, the data first passes through the CBAM module for feature enhancement and then passes through the CFBLS module for feature processing and classification.
[0167] The input transformer data is processed and prediction results are output. These results are usually a set of probability values, indicating the likelihood that the input data belongs to each fault category.
[0168] According to the probability value output by the model, the category with the highest probability is selected as the prediction result. This category is the most likely transformer fault type considered by the model.
[0169] Based on the prediction results, appropriate repair or maintenance decisions are made. For example, if the prediction results indicate a certain fault in the transformer, maintenance personnel may need to be arranged for on-site inspection and repair.
[0170] like Figure 2 As shown, the embodiment of the present application provides a transformer fault diagnosis device, including a signal acquisition module 201, a feature image module 202, a feature matrix module 203, a model construction module 204, a model training module 205 and a fault diagnosis module 206, wherein:
[0171] The signal acquisition module 201 is configured to acquire vibration signals of the transformer core and windings;
[0172] The feature image module 202 is configured to process the vibration signal through the Gram angle field to obtain a feature image, wherein the feature image includes a Gram angle sum field feature image and a Gram angle difference field feature image;
[0173] A feature matrix module 203 is configured to process the Gram angle sum field feature image and the Gram angle difference field feature image through a multi-attention mechanism to obtain a global feature matrix;
[0174] A model building module 204 is configured to perform feature node cascading on the global feature matrix based on the width learning system to build a CFBLS model;
[0175] The model training module 205 is configured to construct a data set based on the feature image and divide the data set into a training set and a test set; input the global feature matrix into the CFBLS model, adjust the parameters of the CFBLS model using the training set, construct a CBAM-CFBLS transformer fault diagnosis model, and test the CBAM-CFBLS transformer fault diagnosis model using the test set until the fault diagnosis accuracy reaches a preset threshold;
[0176] The fault diagnosis module 206 is configured to perform transformer fault diagnosis using the trained CBAM-CFBLS transformer fault diagnosis model.
[0177] The transformer fault diagnosis device of the embodiment of the present application can be a computer device, or a component in the computer device, such as an integrated circuit or a chip. The computer device can be a terminal, or other devices other than a terminal. For example, the computer device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted computer device, a mobile Internet device (MID), an ultra-mobile personal computer (UMPC), a netbook, or a personal digital assistant (PDA), etc. It can also be a server, a network attached storage (NAS), a personal computer (PC), etc., and the embodiment of the present application does not specifically limit it.
[0178] The transformer fault diagnosis device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented in the transformer fault diagnosis method embodiment are not described here.
[0179] The present application also provides a computer device, such as Figure 3 As shown, the computer device includes a processor 301 and a memory 302. The memory 302 stores a program or instruction that can be run on the processor 301. When the program or instruction is executed by the processor 301, each step of the above-mentioned transformer fault diagnosis method is implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0180] It should be noted that the computer devices in the embodiments of the present application include the above-mentioned mobile computer devices and non-mobile computer devices.
[0181] The memory 302 can be used to store software programs and various data. The memory 302 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store an operating system, applications or instructions required for at least one function (such as a sound playback function, an image playback function, etc.). In addition, the memory 302 may include volatile memory or non-volatile memory, or the memory 302 may include both volatile and non-volatile memory. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DRRAM). The memory 302 of the embodiment of the present application includes but is not limited to these and any other suitable types of memory.
[0182] Processor 301 may include one or more processing units. Optionally, processor 301 integrates an application processor and a modem processor. The application processor primarily handles operations related to the operating system, user interface, and application programs, while the modem processor primarily processes wireless communication signals, such as a baseband processor. It is understood that the modem processor may not be integrated into processor 301.
[0183] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned transformer fault diagnosis method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.
[0184] An embodiment of the present application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above-mentioned transformer fault diagnosis method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0185] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.
[0186] An embodiment of the present application also provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the various processes of the above-mentioned transformer fault diagnosis method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.
[0187] It should be noted that, in the present application, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.
[0188] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.
Claims
1. A transformer fault diagnosis method, characterized in that: The method comprises: Obtain vibration signals of transformer core and winding; Processing the vibration signal through a Gram angle field to obtain a characteristic image, wherein the characteristic image includes a Gram angle sum field characteristic image and a Gram angle difference field characteristic image; The Gram angle sum field feature image and the Gram angle difference field feature image are processed by a multi-attention mechanism to obtain a global feature matrix, including: The Gram angle sum field feature image and the Gram angle difference field feature image are used as inputs of a multi-attention mechanism, which includes a channel attention mechanism and a spatial attention mechanism. The formula of the channel attention mechanism is as follows: Where i = 1, 2, ..., n, Z i ′ represents the features output by the channel attention mechanism of the i-th group of feature maps, M c represents the attention weight in the channel dimension, It represents the feature map obtained after GAF transformation of the input i-th group of vibration signal samples. represents element-wise multiplication; The features output by the channel attention mechanism are used as the input of the spatial attention mechanism. The formula is as follows: Among them, M s represents the attention weight in the spatial dimension; Z i ″ represents the features output by the spatial attention mechanism of the i-th group of feature maps; The formula of the global feature matrix is as follows: Based on the width learning system, the feature nodes of the global feature matrix are cascaded to construct a composite feature width learning system (CFBLS) model, including: The global feature matrix is input into the width learning system. The width learning system converts the global feature matrix into n groups of random feature nodes through linear mapping, and uses the output of the previous group of feature nodes as the input of the next group of feature nodes to obtain the i-th group of feature nodes of the CFBLS model. The formula is as follows: Where i = 1, 2, ..., n, represents the feature layer activation function, W ei and β ei Respectively represent the weight and bias of the randomly generated i-th feature node; Each set of feature nodes of the CFBLS model is connected in series to obtain a complete feature mapping layer Z p , the formula is as follows: WITH p =[Z1,Z2,...,Z p ] For the complete feature map node Z p Perform nonlinear transformation to obtain enhanced nodes. The formula is as follows: H j =ξ(ZW hk +b hk ) Where j = 1, 2, ..., q, ξ(·) represents the linear activation function of the enhancement node, W hk and β hk Represent the weight and bias of the randomly generated k-th enhancement node respectively; Each enhancement node is connected in series to obtain the complete enhancement layer H q , the formula is as follows: H q =[H1H2,...,H q ] The feature map layer Z p and the reinforcing layer H q After being connected in series to the output layer, a complete CFBLS model is formed. The formula of the output matrix is as follows: <h2 style=";text-align:left;direction:ltr">Y=[Z<h2 style=";text-align:left;direction:ltr"> p <h2 style=";text-align:left;direction:ltr"> |H<h2 style=";text-align:left;direction:ltr"> q <h2 style=";text-align:left;direction:ltr"> ]W=AW Where A=[Z p |H q ], represents the hidden layer of all feature mapping layers and enhancement layers, and W represents the weight connecting the hidden layer and the output layer; The objective function for calculating the optimal output weight is as follows: Where λ represents the regularization parameter; The optimal output weight W is calculated by the Tikhonov regularization method best , the formula is as follows: W best =A + Y=(λI+A T A) -1 A T Y Among them, A + represents the pseudo-inverse matrix of A, and I represents the identity matrix; When λ→0, the pseudo-inverse matrix A + The formula is as follows: A data set is constructed based on the feature image, and the data set is divided into a training set and a test set; the global feature matrix is input into the CFBLS model, the parameters of the CFBLS model are adjusted using the training set, a CBAM-CFBLS transformer fault diagnosis model is constructed, and the CBAM-CFBLS transformer fault diagnosis model is tested using the test set until the fault diagnosis accuracy reaches a preset threshold; Transformer fault diagnosis is performed using the trained CBAM-CFBLS transformer fault diagnosis model.
2. The transformer fault diagnosis method according to claim 1, wherein: The step of processing the vibration signal by using the Gram angle field to obtain a characteristic image includes: By the sequence X=(X1,X2,...,X n ) represents the vibration signal, the vibration signal is normalized, the sequence X is normalized to the interval [0,1], and the normalized value is obtained The formula is as follows: Where i = 1, 2, ..., n, x i represents the i-th group of vibration signals, max(X) represents the minimum value in the vibration signal, and min(X) represents the minimum value in the vibration signal.
3. The transformer fault diagnosis method according to claim 2, characterized in that: The method includes mapping the sequence X into a time series in polar coordinates, as shown below: Among them, t i represents the time node, N represents the regularization constant factor in polar coordinates, θ i represents the level angle of point i, r represents the polar coordinate radius, Represents the normalized time series.
4. The transformer fault diagnosis method according to claim 3, characterized in that: The method includes eliminating noise in the vibration signal through two inner products with penalty terms, as shown in the following formula: Here, <·> represents the inner product operation.
5. The transformer fault diagnosis method according to claim 3, characterized in that: The method comprises calculating the cosine function of the sum of every two level angles and summing them up to generate a Gram angle and field two-dimensional feature image, and the formula is as follows: By calculating the sine function of the difference between every two level angles and summing them up, a two-dimensional characteristic image of the Gram angle difference field is generated. The formula is as follows: Where I′ represents the unit row vector, represents the normalized time series, express The transpose of .
6. A transformer fault diagnosis device, characterized in that: The device comprises: A signal acquisition module is configured to acquire vibration signals of the transformer core and winding; a feature image module configured to process the vibration signal through a Gram angle field to obtain a feature image, wherein the feature image includes a Gram angle sum field feature image and a Gram angle difference field feature image; The feature matrix module is configured to process the Gram angle sum field feature image and the Gram angle difference field feature image through a multi-attention mechanism to obtain a global feature matrix, including: The Gram angle sum field feature image and the Gram angle difference field feature image are used as inputs of a multi-attention mechanism, which includes a channel attention mechanism and a spatial attention mechanism. The formula of the channel attention mechanism is as follows: Where i = 1, 2, ..., n, Z i ′ represents the features output by the channel attention mechanism of the i-th group of feature maps, M c represents the attention weight in the channel dimension, It represents the feature map obtained after GAF transformation of the input i-th group of vibration signal samples. represents element-wise multiplication; The features output by the channel attention mechanism are used as the input of the spatial attention mechanism. The formula is as follows: Among them, M s represents the attention weight in the spatial dimension; Z i ″ represents the features output by the spatial attention mechanism of the i-th group of feature maps; The formula of the global feature matrix is as follows: A model building module is configured to perform feature node cascading on the global feature matrix based on the width learning system to build a composite feature broad learning system (CFBLS) model, including: The global feature matrix is input into the width learning system. The width learning system converts the global feature matrix into n groups of random feature nodes through linear mapping, and uses the output of the previous group of feature nodes as the input of the next group of feature nodes to obtain the i-th group of feature nodes of the CFBLS model. The formula is as follows: Where i = 1, 2, ..., n, represents the feature layer activation function, W ei and β ei Respectively represent the weight and bias of the randomly generated i-th feature node; Each set of feature nodes of the CFBLS model is connected in series to obtain a complete feature mapping layer Z p , the formula is as follows: WITH p =[Z1,Z2,...,Z p ] For the complete feature map node Z p Perform nonlinear transformation to obtain enhanced nodes. The formula is as follows: H j =ξ(ZW hk +b hk ) Where j = 1, 2, ..., q, ξ(·) represents the linear activation function of the enhancement node, W hk and β hk Represent the weight and bias of the randomly generated k-th enhancement node respectively; Each enhancement node is connected in series to obtain the complete enhancement layer H q , the formula is as follows: H q =[H1H2,...,H q ] The feature map layer Z p and the reinforcing layer H q After being connected in series to the output layer, a complete CFBLS model is formed. The formula of the output matrix is as follows: <h2 style=";text-align:left;direction:ltr">Y=[Z<h2 style=";text-align:left;direction:ltr"> p <h2 style=";text-align:left;direction:ltr"> |H<h2 style=";text-align:left;direction:ltr"> q <h2 style=";text-align:left;direction:ltr"> ]W=AW Where A=[Z p |H q ], represents the hidden layer of all feature mapping layers and enhancement layers, and W represents the weight connecting the hidden layer and the output layer; The objective function for calculating the optimal output weight is as follows: Where λ represents the regularization parameter; The optimal output weight W is calculated by the Tikhonov regularization method best , the formula is as follows: W best =A + Y=(λI+A T A) -1 A T Y Among them, A + represents the pseudo-inverse matrix of A, and I represents the identity matrix; When λ→0, the pseudo-inverse matrix A + The formula is as follows: a model training module configured to construct a data set based on the feature image and divide the data set into a training set and a test set; input the global feature matrix into the CFBLS model, adjust the parameters of the CFBLS model using the training set, construct a CBAM-CFBLS transformer fault diagnosis model, and test the CBAM-CFBLS transformer fault diagnosis model using the test set until the fault diagnosis accuracy reaches a preset threshold; The fault diagnosis module is configured to perform transformer fault diagnosis using the trained CBAM-CFBLS transformer fault diagnosis model.
7. A transformer fault diagnosis system, comprising a processor and a memory, wherein a computer program is stored in the memory, characterized in that: The computer program is loaded and executed by the processor to implement the transformer fault diagnosis method according to any one of claims 1 to 5.
8. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, it is used to implement the transformer fault diagnosis method according to any one of claims 1 to 5.
Citation Information
Patent Citations
Rolling bearing fault diagnosis method based on GAF-DRSN
CN114595730A
Bearing fault diagnosis method and system based on improved deep residual algorithm
CN114861718A