Fault detection method and device for gas insulated switchgear, and electronic equipment
By constructing a two-dimensional feature matrix of gas-insulated switching equipment and using the Kolmogorov-Arnold neural network model, the problem of low GIS fault detection efficiency is solved, and fast and accurate fault detection and preventive maintenance are achieved.
Patent Information
- Application Number
- CN202510312775.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-14
- Publication Date
- 2025-07-22
AI Technical Summary
In the prior art, gas insulated switchgear (GIS) has low fault detection efficiency and large calculation amount, making it difficult to quickly and accurately detect internal mechanical structure defects, resulting in the development of potential faults into serious consequences.
By obtaining the vibration signals of gas insulated switching equipment, a two-dimensional feature matrix is constructed, and fault detection is performed using the Kolmogorov-Arnold neural network (KAN) model to reduce the calculation amount and improve detection efficiency.
It realizes fast and accurate GIS fault detection, can detect potential problems in advance, reduce the occurrence rate of failure, and ensure the stable operation of the equipment.
Smart Images
Figure CN120354231A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of power transmission, and particularly to a fault detection method, device, and electronic device for a Gas Insulated Switchgear (GIS). Background Art
[0002] Due to its advantages such as strong power transmission capacity, small floor area, and less maintenance work, GIS is widely used in the field of power transmission. However, it is difficult to detect internal mechanical structure defects in GIS equipment, and defects are likely to develop into faults under long-term live operation, resulting in serious consequences.
[0003] In existing research, mainly numerical feature extraction or neural network methods are used to solve the above-mentioned fault detection problem by using a one-dimensional feature vector composed of multiple features.
[0004] However, the existing methods have disadvantages such as a large computational amount of the network model, resulting in low detection efficiency and a complicated detection process. Summary of the Invention
[0005] Embodiments of this application provide a fault detection method, device, and electronic device for a gas insulated switchgear, so as to achieve efficient and convenient fault detection of GIS.
[0006] In a first aspect, embodiments of this application provide a fault detection method for a gas insulated switchgear, including:
[0007] Obtain a target vibration signal of the gas insulated switchgear to be detected;
[0008] Construct a target two-dimensional feature matrix corresponding to the target vibration signal;
[0009] Input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas insulated switchgear to be detected, where the fault detection model is trained based on at least one vibration signal of the gas insulated switchgear in different mechanical states for a Kolmogorov–Arnold neural network.
[0010] In a possible implementation manner, the constructing a target two-dimensional feature matrix corresponding to the target vibration signal includes:
[0011] Segment the target vibration signal with a steady-state vibration period to obtain at least one first feature vector corresponding to a target vibration segment;
[0012] Concatenate the at least one first feature vector corresponding to the target vibration segment in chronological order with a preset step length to construct the target two-dimensional feature matrix.
[0013] In a possible implementation manner, before inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected, the method further includes:
[0014] Obtain at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time period;
[0015] For each mechanical state, construct a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state;
[0016] Train the Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain the fault detection model of the gas-insulated switchgear.
[0017] In a possible implementation manner, the constructing a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state includes:
[0018] Segment the vibration signal with a steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment;
[0019] Stitch the second feature vectors corresponding to the at least one vibration segment in chronological order with a preset step length to construct the two-dimensional feature matrix.
[0020] In a possible implementation manner, the segmenting the vibration signal with a steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment includes:
[0021] Segment the vibration signal with the steady-state vibration period to obtain the at least one vibration segment;
[0022] For each vibration segment, perform waveform quantization description on the vibration segment to obtain a second feature vector corresponding to the vibration segment.
[0023] In a possible implementation manner, the training the Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain the fault detection model of the gas-insulated switchgear includes:
[0024] For each mechanical state, divide the mechanical state and the two-dimensional feature matrix corresponding to the mechanical state into a training set and a validation set according to a preset ratio;
[0025] Train the Kolmogorov–Arnold neural network according to the training set and the validation set until the detection performance reaches a preset value to obtain the fault detection model.
[0026] In a possible implementation manner, the second feature vector includes at least one of the following waveform parameters: vibration mean value, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, and frequency variance.
[0027] In a second aspect, an embodiment of the present application provides a fault detection device for a gas-insulated switchgear, including:
[0028] An acquisition module, configured to acquire a target vibration signal of a gas-insulated switchgear to be detected;
[0029] A construction module, configured to construct a target two-dimensional feature matrix corresponding to the target vibration signal;
[0030] A processing module, configured to input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain a target mechanical state of the gas-insulated switchgear to be detected, where the fault detection model is trained based on at least one vibration signal of the gas-insulated switchgear in different mechanical states for a Kolmogorov–Arnold neural network.
[0031] In a possible implementation manner, the construction module is specifically configured to:
[0032] Segment the target vibration signal according to a steady-state vibration period to obtain a first feature vector corresponding to at least one target vibration segment;
[0033] Stitch the first feature vectors corresponding to the at least one target vibration segment in chronological order according to a preset step length to construct the target two-dimensional feature matrix.
[0034] In a possible implementation manner, before inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain a target mechanical state of the gas-insulated switchgear to be detected, a training module is configured to:
[0035] Acquire at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time period;
[0036] For each mechanical state, construct a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state;
[0037] Train the Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain a fault detection model of the gas-insulated switchgear.
[0038] In a possible implementation manner, when the training module constructs a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state, it is specifically configured to:
[0039] Segment the vibration signal by the steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment;
[0040] Stitch together the second feature vectors corresponding to the at least one vibration segment in chronological order with a preset step size to construct the two-dimensional feature matrix.
[0041] In a possible implementation manner, the training module segments the vibration signal by the steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment, and specifically is used for:
[0042] Segment the vibration signal by the steady-state vibration period to obtain the at least one vibration segment;
[0043] For each vibration segment, perform waveform quantification description on the vibration segment to obtain a second feature vector corresponding to the vibration segment.
[0044] In a possible implementation manner, the training module trains a Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain a fault detection model for the gas-insulated switchgear, and specifically is used for:
[0045] For each mechanical state, divide the mechanical state and the two-dimensional feature matrix corresponding to the mechanical state into a training set and a validation set according to a preset ratio;
[0046] Train the Kolmogorov–Arnold neural network according to the training set and the validation set until the detection performance reaches a preset value to obtain the fault detection model.
[0047] In a possible implementation manner, the second feature vector includes at least one of the following waveform parameters: vibration mean, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, frequency variance.
[0048] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;
[0049] The memory stores computer execution instructions;
[0050] The processor executes the computer execution instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementation manners of the first aspect.
[0051] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored, and when the computer execution instructions are executed by a processor, they are used to implement the above first aspect and / or various possible implementation manners of the first aspect.
[0052] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which when executed by a processor implements the above first aspect and / or various possible implementation manners of the first aspect.
[0053] The fault detection method, device, and electronic device for a gas-insulated switchgear provided by the embodiments of the present application obtain a target vibration signal of the gas-insulated switchgear to be detected; construct a target two-dimensional feature matrix corresponding to the target vibration signal; input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected. The fault detection model is trained based on at least one vibration signal of the gas-insulated switchgear in different mechanical states for a Kolmogorov–Arnold neural network. In this technical solution, by converting the vibration signal of the gas-insulated switchgear to be detected into a two-dimensional feature matrix to reflect the time-frequency related features of the device vibration, the time-series related calculation amount in subsequent model processing is reduced. Then, using the Kolmogorov–Arnold neural network, by performing feature representation on the two-dimensional feature matrix, the mechanical state of the gas-insulated switchgear to be detected can be efficiently obtained to quickly and accurately reflect the faults existing in the device. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present application and used together with the specification to explain the principles of the present application.
[0055] Figure 1 Schematic flowchart of fault detection for the gas-insulated switchgear provided by the embodiments of the present application Figure 1 ;
[0056] Figure 2 Schematic diagram of the construction process of the target two-dimensional feature matrix provided by the embodiments of the present application;
[0057] Figure 3 Schematic flowchart of fault detection for the gas-insulated switchgear provided by the embodiments of the present application Figure 2 ;
[0058] Figure 4 Schematic diagram of the confusion matrix of the classification result provided by the embodiments of the present application;
[0059] Figure 5 Schematic flowchart of fault detection for the gas-insulated switchgear provided by the embodiments of the present application Figure 3 ;
[0060] Figure 6 Schematic diagram of the structure of the fault detection device for the gas-insulated switchgear provided by the embodiments of the present application;
[0061] Figure 7 This is a schematic structural diagram of the electronic device provided by the embodiment of the present application.
[0062] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and more detailed descriptions will be provided hereinafter. These drawings and written descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed implementation manners
[0063] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0064] Due to its advantages such as strong power transmission capacity, small floor area, and less maintenance work, GIS is widely used in the field of power transmission. However, it is difficult to detect mechanical structure defects inside GIS equipment, and defects are likely to develop into faults under the long-term live operation state, resulting in serious consequences. Since GIS mechanical structure defects have characteristics such as weak features, strong latency, and unclear features, the research on mechanical defect diagnosis for GIS equipment has attracted much attention.
[0065] In the prior art, machine learning can automatically adjust structural parameters, change its mapping relationship, and has strong generalization ability, and has been widely used in the field of fault diagnosis. For example, the feature vectors of GIS acoustic wave data are extracted, and a fault diagnosis algorithm based on an online sequential extreme learning machine is adopted to achieve accurate diagnosis of mechanical defects.
[0066] However, in the above implementation, methods of numerical feature extraction or neural networks are mainly used, and a one-dimensional feature vector composed of multiple features is used to solve the above-mentioned fault detection problem, which has disadvantages such as a large computational amount of the network model, resulting in low detection efficiency and complicated detection process.
[0067] Based on the above existing technical problems, the technical concept of the inventor is as follows: Existing research mainly uses methods of numerical feature extraction or neural networks. The one-dimensional feature vectors composed of multiple features lack time information, and these methods have a large computational amount. To avoid these problems, a matrix reflecting the change characteristics of relevant features of vibration signals over multiple periods can be constructed. To reduce the computational amount, the Kolmogorov–Arnold Networks (KAN) model is used as the defect classification model. This method can effectively achieve mechanical defect classification based on the feature matrix. Specifically, several vibration periods can be selected from the collected GIS vibration signals, and various mechanical vibration features are extracted and spliced into a multi-mechanical feature vibration vector. Secondly, in chronological order, the multi-mechanical feature vibration vectors are spliced together to construct a multi-dimensional mechanical vibration feature matrix. On this basis, the KAN model is used to perform feature dimensionality reduction on the multi-dimensional mechanical vibration feature matrix and obtain the diagnostic classification result.
[0068] The following uses specific embodiments to elaborate in detail on the technical solutions of the present application and how the technical solutions of the present application solve the above technical problems. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following will describe the embodiments of the present application in conjunction with the drawings.
[0069] Figure 1 Flow schematic diagram of fault detection for gas-insulated switchgear provided by an embodiment of the present application Figure 1 , as Figure 1 shown, the method includes:
[0070] Step 11: Obtain the target vibration signal of the gas-insulated switchgear to be detected;
[0071] In this step, the vibration signal of the gas-insulated switchgear to be detected within a preset time duration is collected and recorded as the target vibration signal.
[0072] In a possible implementation, a piezoelectric acceleration sensor can be used for collection, and the execution entity of the embodiment of the present application obtains it from the sensor.
[0073] Furthermore, the collected target vibration signal can be processed by median filtering to remove noise and interference components and improve the accuracy of subsequent detection.
[0074] It should be understood that the preset time duration here can be a period of time corresponding to the actual detection process.
[0075] Step 12: Construct a target two-dimensional feature matrix corresponding to the target vibration signal;
[0076] In this step, since the amount of data of the target vibration signal of the gas-insulated switchgear to be detected is huge, waveform quantitative description is required.
[0077] That is, the eigenvector of the target vibration signal obtained from a single vibration cycle within the above-mentioned preset time duration has a certain feature representation ability. However, this eigenvector only reflects the features within this cycle and cannot reflect the change characteristics of the vibration signal features among multiple cycles. To solve the limitation of the one-dimensional eigenvector in expression ability in the prior art, in this step, the analysis of the change of eigenvectors between period segments is introduced. By splicing, the one-dimensional eigenvector is converted into a two-dimensional feature matrix based on multiple cycles, that is, the target two-dimensional feature matrix, which can also be called a multi-dimensional mechanical vibration feature matrix.
[0078] Optionally, a possible implementation of step 12 can be:
[0079] Figure 2 For the schematic diagram of the construction process of the target two-dimensional feature matrix provided by the embodiment of the present application, in combination with Figure 2 , step 12 will be described in detail.
[0080] Step 1: Segment the target vibration signal with a steady-state vibration period to obtain at least one first eigenvector corresponding to a target vibration segment;
[0081] In this implementation, by performing segmentation processing on the target vibration signal, at least one target vibration segment can be obtained. Then, by performing waveform quantitative description on each segmented target vibration segment, the eigenvector corresponding to this target vibration segment can be obtained, denoted as the first eigenvector.
[0082] For example, the steady-state vibration period can be the vibration period under power frequency current excitation, such as 0.01 s. As the segmentation length, the target vibration signal within a preset time duration, such as 1 s, is segmented, and 100 target vibration segments can be obtained. For each target vibration segment, waveform quantitative description is performed to obtain at least one waveform parameter among vibration mean, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, and frequency variance. Then, at least one waveform parameter combination is used as a one-dimensional eigenvector, denoted as the first eigenvector.
[0083] Step 2: Splice the first eigenvectors corresponding to at least one target vibration segment in chronological order with a preset step length to construct a target two-dimensional feature matrix.
[0084] In this implementation, sampling is performed in the manner of a fixed-step sliding window. Using the preset step length as the moving step length and the steady-state vibration period as the window length, the one-dimensional vibration feature vectors within the above-mentioned multiple target vibration segments are vectorially concatenated in chronological order to obtain a two-dimensional mechanical vibration feature matrix, that is, the target two-dimensional feature matrix. The column vectors of this matrix represent the vibration features within the period, and the row vectors represent the changes in the feature vectors between the period segments, as Figure 2 shown. The two-dimensional feature matrix of this multiple-period information can more comprehensively reflect the dynamic change characteristics of the vibration signal in multiple periods.
[0085] For example, the preset step length can be 0.0025 s.
[0086] Among them, the time-domain vibration signal is the target vibration signal within the preset duration.
[0087] Step 13: Input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected.
[0088] Among them, the fault detection model is obtained by training the Kolmogorov–Arnold neural network based on at least one vibration signal of the gas-insulated switchgear in different mechanical states.
[0089] In this step, after obtaining the target two-dimensional feature matrix, input the target two-dimensional feature matrix into a pre-trained fault detection model. The training process of this fault detection model can be given by the following embodiments, and then output the mechanical state of the gas-insulated switchgear to be detected, that is, the target mechanical state.
[0090] Among them, the mechanical state can include various possible states such as the GIS device being in a normal state, the bellows bolt being loose, and foreign objects in the bus gas chamber bolts.
[0091] The fault detection method for gas-insulated switchgear provided by the embodiments of the present application includes obtaining the target vibration signal of the gas-insulated switchgear to be detected; constructing the target two-dimensional feature matrix corresponding to the target vibration signal; inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected. The fault detection model is obtained by training the Kolmogorov–Arnold neural network based on at least one vibration signal of the gas-insulated switchgear in different mechanical states. In this technical solution, by converting the vibration signal of the gas-insulated switchgear to be detected into a two-dimensional feature matrix to reflect the time-frequency related features of the device vibration, the time-series related calculation amount in subsequent model processing is reduced. Then, using the Kolmogorov–Arnold neural network, by performing feature characterization on the two-dimensional feature matrix, the mechanical state of the gas-insulated switchgear to be detected can be efficiently obtained to quickly and accurately reflect the faults existing in the device.
[0092] Based on the above embodiments, Figure 3 The following is a schematic flow chart of the fault detection of the gas-insulated switchgear provided in the embodiments of the present application. Figure 2 As Figure 3 shown, before step 13, the method further includes:
[0093] Step 31: Obtain at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time period;
[0094] In this step, during the training process, the gas-insulated switchgear in different mechanical states can be respectively collected for a preset time period, so as to obtain the vibration signals respectively corresponding to different mechanical states.
[0095] For example, a piezoelectric acceleration sensor is used to collect the vibration signals on the shell of the GIS equipment in three mechanical states: normal, bellows bolt loose (defect 1), and bus chamber bolt foreign object (defect 2).
[0096] Furthermore, the collected vibration signals can also be filtered by means of median filtering.
[0097] Step 32: For each mechanical state, construct a two-dimensional feature matrix according to the vibration signal corresponding to the mechanical state;
[0098] Optionally, the implementation of step 32 can be:
[0099] Step 1: Segment the vibration signal with the steady-state vibration period to obtain at least one second feature vector corresponding to the vibration segment;
[0100] In a possible implementation, it can be: segment the vibration signal with the steady-state vibration period to obtain at least one vibration segment; for each vibration segment, perform waveform quantification description on the vibration segment to obtain the second feature vector corresponding to the vibration segment.
[0101] Exemplarily, the second feature vector includes at least one of the following waveform parameters: vibration mean value, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, frequency variance.
[0102] Step 2: Concatenate the second feature vectors corresponding to at least one vibration segment in chronological order with a preset step length to construct a two-dimensional feature matrix.
[0103] It should be understood that the implementation process of this step 32 and the implementation process of step 12 and Figure 2 the example are similar. For the parts not described in detail, reference can be made to the above implementation.
[0104] Step 33: Train a Kolmogorov–Arnold neural network based on each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain a fault detection model for the gas-insulated switchgear.
[0105] In this step, the neural network model constructed by the Kolmogorov–Arnold representation theorem can efficiently approximate multi-dimensional continuous functions. By decomposing a high-dimensional function (i.e., the two-dimensional feature matrix) into a combination of multiple univariate functions, the network structure is simplified and the approximation accuracy is improved.
[0106] Among them, the high-dimensional function f(x) can be expressed as a combination of a finite number of univariate continuous functions, and the principle expression is as follows:
[0107]
[0108] Among them, the value range of φ q,p is from 0 to 1, the value range of φ q is all real numbers, n is the number of feature vectors in the two-dimensional feature matrix. Traditional neural networks use fixed activation functions, while KAN introduces learnable activation functions at the edges of the network. This design enables each weight parameter in KAN to be replaced by a univariate function, and these functions are usually parameterized in the form of spline functions, thus endowing the network with extremely high flexibility. In this way, KAN can accurately approximate complex functions with fewer parameters, enhancing the interpretability of the model. The mechanical defect model used in this patent consists of an input layer, a hidden layer, and a judgment layer, where the number of neurons is 224, 224, and 3 respectively.
[0109] Compared with the method of traditional Convolutional Neural Network (CNN), it has the advantage of less computational complexity. The floating point operations (FLOPs) of a network are often used to describe the complexity of a network and can be understood as the amount of computation. Here, ResNet18 is used as the representative algorithm of CNN to compare the differences between the two algorithms and show the advantages of this solution, that is, the comparison of the network FLOPs and the number of model parameters of KAN and CNN is shown in Table 1.
[0110] Table 1
[0111] Model Model FLOPs / Number of model parameters ResNet18 1.83*109 Mac 11.44M KAN 448 Mac 508.48k
[0112] It can be seen that in the horizontal comparison of neural network methods, the KAN model has the advantage of less computational complexity compared with the traditional CNN model, which is convenient for the model to be deployed in on-site edge devices with poor computing power.
[0113] Optionally, the implementation of step 33 can be:
[0114] Exemplarily, Figure 4 This is a schematic diagram of the confusion matrix of the classification results (i.e., mechanical states) provided by the embodiments of the present application. In combination with Figure 4 , step 33 will be described in detail.
[0115] Step 1: For each mechanical state, divide the mechanical state and the two-dimensional feature matrix corresponding to the mechanical state into a training set and a validation set according to a preset ratio;
[0116] In this implementation, taking the 550 kV full-scale gas-insulated switchgear as an example, the current sharing capacity is set. With a gradient of 1000 A, in the range from 1000 A to 5000 A, different magnitudes of current are applied to simulate two common mechanical defects, namely bellows bolt loosening (defect 1) and bus chamber bolt foreign object (defect 2), and a normal state is set, etc., for a total of 3 mechanical states.
[0117] The collected GIS vibration signals under different mechanical states and different current-carrying conditions are divided into a training set (e.g., 2250) and a test set (e.g., 750) according to a preset ratio, such as a ratio of 3:1, to establish a multi-mechanical vibration feature matrix of normal signals and two defect vibration signals, and use it as an input to the KAN network.
[0118] Step 2: Train the Kolmogorov–Arnold neural network according to the training set and the validation set until the detection performance reaches a preset value to obtain a fault detection model.
[0119] In this implementation, based on the training set and the validation set, the above-mentioned Kolmogorov–Arnold neural network is trained. This preset value can be used as a criterion for testing the training performance of the model. For example, the accuracy rate, recall rate, and F1-score (i.e., F1) used as evaluation indicators are all greater than 93%.
[0120] For example, the accuracy rate A cc , the recall rate R call , and the formula for F1 is as follows:
[0121]
[0122] Among them, TN and TP are correctly identified normal and fault samples, and FN and FP are misidentified as normal and fault samples.
[0123] After the above input to the KAN network and training, the confusion matrix of the classification results during verification is as Figure 4 shown. From the results, it can be seen that the accuracy rate, precision rate, and F1 score are all above 93%. The proposed diagnostic method is reliable, that is, a fault detection model is obtained.
[0124] The fault detection method for gas-insulated switchgear provided by the embodiments of the present application obtains at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time duration. For each mechanical state, a two-dimensional feature matrix in the mechanical state is constructed according to the vibration signal corresponding to the mechanical state. Based on each mechanical state and the corresponding two-dimensional feature matrix, a Kolmogorov–Arnold neural network is trained to obtain a fault detection model for the gas-insulated switchgear. In this technical solution, by obtaining the vibration signals of the gas-insulated switchgear in different mechanical states within a preset time duration and constructing a corresponding two-dimensional feature matrix for each mechanical state, the key time-frequency features of the equipment vibration signals can be effectively extracted. By using each mechanical state and its corresponding two-dimensional feature matrix as inputs to train the Kolmogorov–Arnold neural network, a high-precision fault detection model can be obtained. This model can accurately identify the changes in the mechanical state of the equipment and achieve rapid fault diagnosis based on the deep features of the vibration signals, effectively improving the accuracy, reliability, and real-time performance of fault detection, helping to detect potential problems of the equipment in advance, thereby realizing preventive maintenance, reducing the failure rate, and ensuring the stable operation of the equipment.
[0125] Based on the above embodiments, Figure 5 is a schematic flowchart of the fault detection of the gas-insulated switchgear provided by the embodiments of the present application Figure 3 , as Figure 5 shown, the training and verification of the model may include:
[0126] Step 51: Acquisition and filtering of vibration signals of the GIS equipment in various mechanical states;
[0127] Step 52: Establishment of a multi-dimensional mechanical vibration characteristic matrix of the GIS equipment;
[0128] Step 53: Construction of a KAN mechanical defect identification model for the GIS equipment;
[0129] Step 54: Verification of the reliability of the mechanical defect diagnosis method for the GIS equipment.
[0130] The fault detection method for gas-insulated switchgear provided by the embodiments of the present application has a similar implementation principle and technical effect to the above embodiments, which will not be elaborated here.
[0131] The following is an explanation of the device embodiments provided by the embodiments of the present application.
[0132] Figure 6 is a schematic structural diagram of the fault detection device for gas-insulated switchgear provided by the embodiments of the present application. As Figure 6 shown, the fault detection device for gas-insulated switchgear provided in this embodiment includes:
[0133] An acquisition module 61 for acquiring a target vibration signal of a gas-insulated switchgear to be detected;
[0134] A construction module 62 for constructing a target two-dimensional feature matrix corresponding to the target vibration signal;
[0135] A processing module 63 for inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected, where the fault detection model is trained based on at least one vibration signal of the gas-insulated switchgear in different mechanical states for a Kolmogorov–Arnold neural network.
[0136] In a possible implementation manner, the construction module 62 is specifically configured to:
[0137] Segment the target vibration signal with a steady-state vibration period to obtain a first feature vector corresponding to at least one target vibration segment;
[0138] Stitch the first feature vectors corresponding to at least one target vibration segment in chronological order with a preset step length to construct a target two-dimensional feature matrix.
[0139] In a possible implementation manner, before inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected, a training module is used to:
[0140] Acquire at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time period;
[0141] For each mechanical state, construct a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state;
[0142] Train a Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain a fault detection model of the gas-insulated switchgear.
[0143] In a possible implementation manner, when the training module constructs a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state, it is specifically configured to:
[0144] Segment the vibration signal with a steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment;
[0145] Stitch the second feature vectors corresponding to at least one vibration segment in chronological order with a preset step length to construct a two-dimensional feature matrix.
[0146] In a possible implementation, the training module segments the vibration signal at the steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment, specifically for:
[0147] Segment the vibration signal at the steady-state vibration period to obtain at least one vibration segment;
[0148] For each vibration segment, perform waveform quantization description on the vibration segment to obtain a second feature vector corresponding to the vibration segment.
[0149] In a possible implementation, the training module trains a Kolmogorov–Arnold neural network based on each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain a fault detection model for the gas-insulated switchgear, specifically for:
[0150] For each mechanical state, divide the mechanical state and the two-dimensional feature matrix corresponding to the mechanical state into a training set and a validation set according to a preset ratio;
[0151] Train the Kolmogorov–Arnold neural network according to the training set and the validation set until the detection performance reaches a preset value to obtain a fault detection model.
[0152] In a possible implementation, the second feature vector includes at least one of the following waveform parameters: vibration mean, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, frequency variance.
[0153] The fault detection device for the gas-insulated switchgear provided in this embodiment can execute the fault detection method for the gas-insulated switchgear provided in the above method embodiment, and its implementation principle and technical effect are similar, so details are not described here in this embodiment.
[0154] Figure 7 This is a schematic structural diagram of an electronic device provided in an embodiment of the present application. As Figure 7 shown, the electronic device provided in this embodiment includes:
[0155] At least one processor 71 and a memory 72.
[0156] Optionally, the device further includes a communication component 73. Among them, the processor 71, the memory 72, and the communication component 73 are connected through a bus 74.
[0157] In a specific implementation process, at least one processor 71 executes computer-executable instructions stored in the memory 72, so that at least one processor 71 executes the above method.
[0158] The specific implementation process of the processor 71 can be referred to the above method embodiment, and its implementation principle and technical effect are similar, so details are not described here again in this embodiment.
[0159] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU for short), or other general-purpose processors, digital signal processors (DSP for short), application specific integrated circuits (ASIC for short), etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the method disclosed in combination with the invention can be directly implemented by the execution of the hardware processor, or implemented by the combination of the hardware and software modules in the processor.
[0160] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.
[0161] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, the buses in the drawings of this application are not limited to only one bus or one type of bus.
[0162] This application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.
[0163] This application also provides a computer-readable storage medium, in which computer-executable instructions are stored, and when the processor executes the computer-executable instructions, the above method is implemented.
[0164] The above-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk or an optical disc. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.
[0165] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an Application Specific Integrated Circuits (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in a device.
[0166] The division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Additionally, the couplings, direct couplings, or communication connections shown or discussed among each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in electrical, mechanical, or other forms.
[0167] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0168] Furthermore, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.
[0169] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.
[0170] Those of ordinary skill in the art will understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disks or optical discs that can store program codes.
[0171] Finally, it should be noted that: After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present invention. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not disclosed by the present invention. It is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.
Claims
1. A fault detection method for a gas-insulated switchgear, characterized in that, Including: Obtain the target vibration signal of the gas-insulated switchgear to be detected; Construct a target two-dimensional feature matrix corresponding to the target vibration signal; Input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected. The fault detection model is trained based on at least one vibration signal of the gas-insulated switchgear in different mechanical states to train the Kolmogorov–Arnold neural network.
2. The method according to claim 1, characterized in that The constructing the target two-dimensional feature matrix corresponding to the target vibration signal includes: Segment the target vibration signal with a steady-state vibration period to obtain a first feature vector corresponding to at least one target vibration segment; Stitch the first feature vectors corresponding to the at least one target vibration segment in chronological order with a preset step length to construct the target two-dimensional feature matrix.
3. The method according to claim 1, characterized in that, Before the inputting the target two-dimensional feature matrix into a pre-trained fault detection model to obtain the target mechanical state of the gas-insulated switchgear to be detected, the method further includes: Obtain at least one vibration signal of the gas-insulated switchgear in different mechanical states within a preset time period; For each mechanical state, construct a two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state; Train the Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain the fault detection model of the gas-insulated switchgear.
4. The method according to claim 3, characterized in that The constructing the two-dimensional feature matrix in the mechanical state according to the vibration signal corresponding to the mechanical state includes: Segment the vibration signal with a steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment; Stitch the second feature vectors corresponding to the at least one vibration segment in chronological order with a preset step length to construct the two-dimensional feature matrix.
5. The method according to claim 4, wherein The segmenting the vibration signal with a steady-state vibration period to obtain a second feature vector corresponding to at least one vibration segment includes: Segment the vibration signal with the steady-state vibration period to obtain the at least one vibration segment; For each vibration segment, perform waveform quantification description on the vibration segment to obtain a second feature vector corresponding to the vibration segment.
6. The method according to any one of claims 3-5, characterized in that, The training the Kolmogorov–Arnold neural network according to each mechanical state and the two-dimensional feature matrix corresponding to each mechanical state to obtain the fault detection model of the gas-insulated switchgear includes: For each mechanical state, divide the mechanical state and the two-dimensional feature matrix corresponding to the mechanical state into a training set and a validation set according to a preset ratio; Train the Kolmogorov–Arnold neural network according to the training set and the validation set until the detection performance reaches a preset value to obtain the fault detection model.
7. The method according to claim 4 or 5, characterized in that, The second feature vector includes at least one of the following waveform parameters: vibration mean, standard deviation, skewness, kurtosis, center frequency, average frequency, root mean square frequency, frequency variance; At least one mechanical state includes at least one of the following: bellows bolt loose, bus chamber bolt foreign object, normal state.
8. A fault detection device for a gas-insulated switchgear, characterized in that, Including: An acquisition module, configured to acquire a target vibration signal of a gas-insulated switchgear to be detected; A construction module, configured to construct a target two-dimensional feature matrix corresponding to the target vibration signal; A processing module, configured to input the target two-dimensional feature matrix into a pre-trained fault detection model to obtain a target mechanical state of the gas-insulated switchgear to be detected, where the fault detection model is obtained by training a Kolmogorov–Arnold neural network based on at least one vibration signal of the gas-insulated switchgear in different mechanical states.
9. An electronic device, characterized in that, Comprising: A memory and a processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor executes the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, Computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, they are used to implement the method according to any one of claims 1-7.