High-voltage circuit breaker abnormity identification method based on vibration signal envelope line

Abstract

The invention provides a high-voltage circuit breaker abnormity identification method based on a vibration signal envelope line. The method comprises the following steps: acquiring an operation vibration signal of a high-voltage circuit breaker, generating vibration data by a conditional GAN, filtering the vibration signal by wavelet transform, extracting the envelope line of the vibration signal and associating an operation state label to form a sample library; an AM-CNN-based high-voltage circuit breaker abnormity identification model is constructed, and sample library data is used for training so as to predict the operation state of the circuit breaker on line; and perfecting the sample library according to the prediction state and updating the circuit breaker abnormity identification model. According to the method, the fault state of the high-voltage circuit breaker is diagnosed by automatically extracting the fault features of the vibration signals, the error state and the unknown fault state are judged through self-learning so as to continuously improve the abnormal recognition precision of the circuit breaker, power transformation operation and maintenance personnel are assisted to make decisions, and the operation state recognition efficiency of the high-voltage circuit breaker is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of high-voltage circuit breaker anomaly recognition, and particularly relates to a high-voltage circuit breaker anomaly recognition method based on vibration signal envelope characteristics. BACKGROUND

[0002] In normal or fault operation of a power system, a high-voltage circuit breaker, as a key device, controls power equipment and line to be put in or cut out. The importance of the high-voltage circuit breaker makes many scholars at home and abroad study the reliability of the circuit breaker. In order to monitor the operating state of the high-voltage circuit breaker, vibration, current and sound signals are often used as monitoring quantities. The sound is easily disturbed by the external environment. The installation of the current sensor is relatively difficult. The vibration sensor, as a non-intrusive fault monitoring device, is favored by scholars. For the collected vibration signals, the time domain, frequency domain and time-frequency domain characteristics of the vibration are often extracted, and then knowledge and machine learning algorithms are used to judge the state of the circuit breaker. However, since the vibration signal of the high-voltage circuit breaker is non-periodic and nonlinear, it is difficult to distinguish the difference between different faults by using a time domain method such as assignment at a certain time. The vibration characteristics are extracted by the frequency domain method of fast Fourier decomposition, and the frequency spectrum is complex and chaotic, so it is difficult to distinguish different faults by using several frequency characteristics. Therefore, the effect of using the time domain or frequency domain method alone to extract fault characteristic quantities is poor.

[0003] In order to solve the above problems, the personnel in the field usually use the time-frequency domain signal processing method of empirical mode decomposition to extract the fault characteristics of the vibration. Although the extraction effect is good, in the mode decomposition process, the vibration signal is generally decomposed into several intrinsic modes, and then the local of each mode is analyzed and the fault characteristics are extracted. The number of modes needs to be tried and adjusted by people. If it is not properly selected, the fault result diagnosis is inaccurate. SUMMARY

[0004] The purpose of the present application is to solve the problems of poor recognition effect and low recognition reliability caused by relying on manual selection of parameters in the existing high-voltage circuit breaker anomaly recognition technology, and to provide a high-voltage circuit breaker anomaly recognition method based on vibration signal envelope.

[0005] To achieve the above purpose, the technical solution provided by the present application is:

[0006] The high-voltage circuit breaker anomaly recognition method based on vibration signal envelope comprises the following steps:

[0007] Step 1: based on the conditional generative adversarial network model, generate a plurality of vibration signal data of the high-voltage circuit breaker corresponding to the specified operating state; extract the envelope of each vibration signal after filtering the vibration signal, and associate the operating state label to form a sample library;

[0008] The running state includes a normal state, a control loop abnormal fault state, a spring loosening fault state and a transmission mechanism jamming fault state.

[0009] Step 2: Constructing a high-voltage circuit breaker abnormality identification model, the model input is the envelope of the vibration signal and the corresponding running state label, and the model output is the running state of the high-voltage circuit breaker; the high-voltage circuit breaker abnormality identification model includes an input layer, a 1D convolution layer, a 1D pooling layer, a 1D spatial attention layer and a Softmax output layer connected in turn;

[0010] The high-voltage circuit breaker abnormality identification model is trained using the sample library data in step 1;

[0011] Step 3: Obtain the vibration signal of the actual high-voltage circuit breaker, filter and denoise the vibration signal to filter out high-frequency interference components; extract the envelope of the denoised vibration signal;

[0012] Step 4: input the envelope of the vibration signal obtained in step 3 into the trained high-voltage circuit breaker abnormality identification model, and the high-voltage circuit breaker abnormality identification model predicts the running state of the high-voltage circuit breaker; if the running state is normal, the envelope of the vibration signal obtained in step 3 is automatically labeled and the sample library is updated; if the running state is abnormal, step 5 is executed;

[0013] The abnormal state includes an abnormal state and an unknown state.

[0014] Step 5: manually check the high-voltage circuit breaker abnormal state, compare the abnormal state determined by manual maintenance with the abnormal state output by the high-voltage circuit breaker abnormality identification model, if they are consistent, the envelope of the vibration signal obtained in step 3 is automatically labeled and the sample library is updated; if they are inconsistent or unknown, the envelope of the vibration signal obtained in step 3 is manually labeled and the sample library is updated.

[0015] Further, step 1 includes the following steps:

[0016] Step 1.1: Perform high-voltage circuit breaker closing or opening operation, and collect multiple groups of real vibration signals of high-voltage circuit breakers in normal and abnormal states; wherein the abnormal state includes control loop abnormal fault state, spring loosening fault state and transmission mechanism jamming fault state, which is realized by artificial simulation;

[0017] Step 1.2: Construct a generative adversarial network model, learn the distribution characteristics of the real vibration signal data obtained in step 1.1 through the adversarial training of the generator and the discriminator in the generative adversarial network model; and synthesize vibration signal data using the trained generator;

[0018] Step 1.3: filtering and denoising the vibration signal data obtained in step 1.1 and step 1.2 by using wavelet transform;

[0019] Step 1.4: extracting the envelope of the vibration signal after filtering and denoising by using the Envelope function, and making a running state label for the envelope of the vibration signal according to the corresponding running state to form a sample library.

[0020] Further, in step 1.1, the corresponding abnormal state is simulated by the following way:

[0021] Control loop abnormal fault state: reduce the control voltage of the high-voltage circuit breaker operating platform;

[0022] Spring loosening fault state: remove the closing spring in the high-voltage circuit breaker operating mechanism and all the gaskets at the closing spring connection, and then install the closing spring back to the operating mechanism;

[0023] Transmission mechanism jamming fault state: hang a metal weight at the transmission mechanism connection to increase the running resistance of the transmission mechanism.

[0024] Further, after step 5, it further includes: training and optimizing the high-voltage circuit breaker abnormality recognition model by using the updated sample library.

[0025] A computer readable storage medium stores a computer executable program, and the computer executable program is used to implement the high-voltage circuit breaker abnormality recognition method when executed.

[0026] An electronic device for data processing, comprising: one or more processors, and the above-mentioned computer readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the high-voltage circuit breaker abnormality recognition method.

[0027] The advantages of the present application are:

[0028] 1、The high-voltage circuit breaker abnormality recognition method of the present application establishes a high-voltage circuit breaker abnormality recognition model based on AM-CNN, only needs to input the vibration sensor signal of the high-voltage circuit breaker, automatically extracts the envelope feature of the vibration signal and recognizes the abnormal state of the high-voltage circuit breaker by using the high-voltage circuit breaker abnormality recognition model, does not need to extract the feature vector manually, reduces the time consumption of fault recognition, improves the fault diagnosis precision, and solves the problems of low recognition efficiency and unreliable recognition results caused by manually trying modal parameters in the traditional method.

[0029] 2、The application constructs a database, utilizes the generative adversarial network (GAN) to generate the circuit breaker operation vibration data, avoids multiple high-voltage circuit breaker operations to obtain vibration data, and influences the subsequent service life of the high-voltage circuit breaker.

[0030] 3、The application further automatically updates the constructed sample database through the model diagnosis result, optimizes the high-voltage circuit breaker abnormal recognition model by using the updated database, and further improves the generalization ability and fault recognition precision of the model. BRIEF DESCRIPTION OF DRAWINGS

[0031] The above and / or additional aspects and advantages of the application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0032] Figure 1 is a general block diagram of the high-voltage circuit breaker abnormal recognition method of the application;

[0033] Figure 2 is a schematic diagram of the original vibration data of four states in the application, wherein, Figure 2 a is a normal state, Figure 2 b is a control loop abnormal fault state, 2c is a spring loosening fault state, and 2d is a transmission mechanism jamming fault state;

[0034] Figure 3 is a principle schematic diagram of generating vibration data based on conditional GAN in the application;

[0035] Figure 4 is a schematic diagram of vibration envelope data of four states in the application; wherein, Figure 4 a is a normal state, Figure 4 b is a control loop abnormal fault state, 4c is a spring loosening fault state, and 4d is a transmission mechanism jamming fault state;

[0036] Figure 5 is a structure diagram of the AM-CNN high-voltage circuit breaker abnormal recognition model constructed in the application;

[0037] Figure 6 is a schematic diagram of the AM-CNN high-voltage circuit breaker abnormal recognition model training process in the application;

[0038] Figure 7 is a comparison diagram of the diagnosis rates of four states under different fault diagnosis models;

[0039] Figure 8 is a comparison diagram of the accuracy and time consumption of different fault diagnosis methods. DETAILED DESCRIPTION

[0040] Embodiments of the present application are described in detail below, which are exemplary and intended to explain the present application, and cannot be understood as a limitation of the present application.

[0041] Referring to Figure 1 , the embodiment takes 220kV LW-252 type SF6 high voltage circuit breaker (HVCB) as an example to describe the specific process of the abnormal recognition method of high voltage circuit breaker based on vibration signal envelope line of the present application. For the collected vibration signal of the closing process of high voltage circuit breaker, the vibration signal is first filtered by wavelet transform, then the envelope line of the vibration signal is extracted according to Envelope, and then the abnormal recognition model of high voltage circuit breaker based on AM-CNN is constructed, and finally the abnormal state of high voltage circuit breaker is output through the abnormal recognition model. Specifically, it includes the following steps:

[0042] Step 1: Constructing a high voltage circuit breaker sample database.

[0043] Step 1.1: The data acquisition system collects part of the real vibration signal data of the high voltage circuit breaker.

[0044] The data acquisition system includes a vibration sensor and a data acquisition card, the data acquisition card uses MP4221, and the vibration sensor is collected at a sampling frequency of 100kHz. When the high voltage circuit breaker has closing action or opening action, the vibration signal amplitude exceeds the sampling threshold, and the data acquisition card continuously records 100ms vibration data.

[0045] In this embodiment, the high voltage circuit breaker closing action is performed, and the high voltage circuit breaker vibration signals under normal state, control loop abnormal fault state, spring loosening fault state and transmission mechanism jamming fault state are collected. The abnormal state is simulated by artificially simulating the circuit breaker fault, and the control voltage of the high voltage circuit breaker operating table is reduced to simulate the control loop abnormal fault state. By removing the closing spring in the operating mechanism of the high voltage circuit breaker, and removing all the gaskets at the spring connection, and then installing the closing spring back to the operating mechanism, the spring loosening fault state is simulated. The transmission mechanism jamming fault state is simulated by hanging a metal weight on the transmission mechanism connection to increase the running resistance of the transmission mechanism. The vibration data of the high voltage circuit breaker under the four states of normal state, control loop abnormal fault, spring loosening fault and transmission mechanism jamming fault are shown in FIG. 2. Figure 2

[0046] According to whether the starting value of the data acquisition card reaches the set starting threshold value, the operation of the circuit breaker is monitored in real time. When the amplitude of the vibration signal reaches the starting threshold value of the data acquisition system, the vibration data of 15 times of closing operation of the circuit breaker under normal state, control loop abnormal fault state, spring loosening fault state and transmission mechanism jamming fault state are collected. The starting value of the data acquisition card in the present application is expressed by speed, and the starting threshold value in this embodiment is set to 5​ .

[0047] Step 1.2: Use a conditional generative adversarial network (GAN) model to generate a plurality of vibration signal data.

[0048] Considering the influence of multiple high-voltage circuit breaker operation vibration data acquisition on the service life, the generative adversarial network (GAN) is used to generate circuit breaker operation vibration data in the application. Figure 3 As shown in the accompanying drawings, the GAN is composed of a generation model and a discrimination model. Suppose there are high-voltage circuit breaker vibration waveforms, and there is a complex distribution relationship between these samples, which is set as Suppose there is a set of noise signals satisfying the joint Gaussian distribution A deep neural network can be used to construct the mapping relationship between and , so that by sampling the known distribution as input, new data satisfying the original data distribution relationship can be generated.

[0049] The training process of GAN is a two-person zero-sum game process, which can be expressed by the formula as follows:

[0050] min G max D V ( G , D ) = E x ~ p r ( x ) [ D ( x )] − E z ~ p z ( z n ) [ D ( G ( z n ))]

[0051] In the formula: represents the expected distribution, is the objective function, is the data generated by the generator, is the output of the discriminator.

[0052] In order to avoid the problems of model training difficulty and mode collapse, the Wasserstein distance is used as the measurement of the two distributions, which is defined as follows:

[0053] W ( p r , p g ) = inf r ~ Pi ( p r , p g ) E ( x , y )~ r [|| x − y ||]

[0054] In the formula: is the set of joint probability distributions and with marginal distribution , is the lower limit of expectation.

[0055] In order to avoid the problems of poor training effect and instability of traditional GAN, the conditional GAN introduces the addition of category information At the input end of the conditional GAN generator, the noise signal combines the condition as the generator Input. Sample It is generated by the generator Output generation. Discriminator of Conditional GAN. The distribution of the generated samples must be determined. Compared with the true sample distribution The degree of similarity, and the generated samples Does it meet the conditions? The optimization objective of conditional GANs, using Wasserstein distance, is as follows:

[0056] min G max D V ( G , D ) = E x ~ p r ( x ) [ D ( x | c )] − E x '~ p z ( x ') [ D ( x '| c )] − lambda w E [|| ∇ D ( epsilon x + ( 1 − epsilon ) G ( z n ))|| − 1 ] 2

[0057] In the formula: It is the coefficient of the regularization term. It is a trade-off factor.

[0058] GAN was used to generate 30 sets of vibration data for each state, so there are a total of 45 sets of vibration data for each state.

[0059] Step 1.3: The vibration signal data obtained in Steps 1.1 and 1.2 are filtered and denoised using wavelet transform. This embodiment employs conventional wavelet transform technology; the specific process is as follows:

[0060] (1) Select the wavelet basis and the number of wavelet decomposition layers to perform multi-scale decomposition on the signal.

[0061] Wavelet basis selection measures the degree of signal disorder by calculating the logarithmic energy entropy after wavelet decomposition. The smaller the entropy value, the more suitable the corresponding wavelet basis is. The formula for calculating the entropy value is:

[0062]

[0063] In the formula, The vibration signal to be analyzed is... It is the sequence number of the vibration signal data point. This refers to the first The wavelet coefficients after wavelet decomposition are first squared and then logarithmic.

[0064] Using signal-to-noise ratio (SNR) as an evaluation criterion for wavelet decomposition level selection, employing an appropriate level to decompose the signal can preserve effective information as much as possible while removing noise. However, excessively large decomposition levels can easily lead to the loss of original information, failing to achieve noise reduction and affecting fault diagnosis results. The expression for SNR is:

[0065]

[0066] In the formula: The signal after denoising and reconstruction. is the original signal.

[0067] After selecting the appropriate wavelet base and decomposition level, the vibration signal is decomposed by wavelet decomposition. The approximation value (low-frequency component) after decomposition is further decomposed by filter to achieve multi-level decomposition. The formula of wavelet decomposition process is:

[0068]

[0069] In the formula, is the approximation value of wavelet decomposition, is the detail value of wavelet decomposition, is the wavelet decomposition level, is the low-pass filter, is the high-pass filter, is the first component of wavelet decomposition, is the approximation value of layer.

[0070] (2) Determine the threshold value and process the wavelet coefficients.

[0071] The selection of threshold value is very important for wavelet denoising. The frequency components of vibration signal are complex. Too large threshold value will remove useful information contained in the vibration signal, and too small threshold value will lead to incomplete noise removal and signal distortion. Common methods for determining threshold value include fixed threshold method, unbiased likelihood estimation threshold method, heuristic threshold method and heuristic threshold method. Any method can be selected to determine the threshold value.

[0072] After selecting the threshold value, the wavelet coefficients are processed. The threshold threshold method is commonly used, including soft threshold method and hard threshold method. The hard threshold method is shown in the following formula:

[0073]

[0074] In the formula, is the original wavelet coefficient; is the threshold value in the threshold function. When the absolute value of is not less than , its value is kept unchanged. When the absolute value of is less than , then is set to 0.

[0075] The soft threshold method is shown in the following formula:

[0076] w lambda = { [ sign ( w ) ] ( | w | − lambda ) | w | ≥ lambda 0 | w | < lambda

[0077] In the formula: when the absolute value of the coefficient after decomposition of the original signal is not less than the threshold value, the coefficient is changed to the original coefficient minus When the coefficient is less than the threshold , the wavelet coefficient is changed to 0.

[0078] (3) Wavelet reconstruction is performed

[0079] Wavelet reconstruction is to use the last layer of the approximate value and the detail value , respectively, to carry out the zero insertion, and then respectively with the filter and convolution operation to obtain layer approximate value , repeat the process, and finally obtain the original signal , expressed by the formula:

[0080]

[0081] After a series of wavelet denoising steps, the purpose of denoising the original signal is achieved.

[0082] Step 1.4: Use the Envelope function to extract the envelope of the vibration signal after filtering. The specific process is as follows:

[0083] The Envelope function is a function for extracting the envelope of the vibration signal, The function returns the upper and lower root mean square envelopes of the denoised vibration signal , and the envelope is determined by the root mean square of the sliding window with a length of samples. Each data point on the envelope is expressed by the formula:

[0084]

[0085] Using the above method, the envelope of each vibration signal is obtained, as shown in the accompanying Figure 4 .

[0086] Step 1.5: Perform data labeling operation to build a sample library.

[0087] According to the envelope corresponding to the operating state of the high-voltage circuit breaker, the data is labeled (0 for normal state, 1 for loop abnormal fault state, 2 for spring loosening fault state, and 3 for transmission mechanism jamming fault state), and a high-voltage circuit breaker opening and closing operation vibration signal sample library is established, which associates the vibration signal envelope with the operating state label.

[0088] Step 2: build an abnormality recognition model of high-voltage circuit breaker based on AM-CNN. The specific process is as follows:

[0089] The AM-CNN-based high-voltage circuit breaker abnormality recognition model built in step 2 is shown in FIG. 2, which includes an input layer, a 1D convolutional layer, a 1D pooling layer, a 1D spatial attention layer, and a Softmax output layer. The envelope of the vibration signal in the database formed in step 1.5 is taken as an input signal of the AM-CNN-based high-voltage circuit breaker abnormality recognition model Figure 5 The label of each operating state is taken as another input signal of the AM-CNN-based high-voltage circuit breaker abnormality recognition model . .

[0090] The 1D convolutional layer includes a plurality of convolutional kernels, each of which slides through the entire envelope of the vibration signal with a certain step to obtain the features of the envelope. The output of the 1D convolutional layer is The mathematical expression is as follows:

[0091]

[0092] In the formula: and are the length and width of the feature map, respectively, is the weight of the one-dimensional convolutional kernel, and are the length and width of the convolutional kernel, respectively, is the convolution operation, is the input data, is the length of the input data, is the bias corresponding to the convolutional kernel.

[0093] In the deep learning algorithm, the activation function is selected as the Relu function, and the mathematical expression is as follows:

[0094]

[0095] The 1D pooling layer is mainly used to reduce the dimension of the selected features of the convolutional layer and has the function of enhancing the features. In CNN, the maximum pooling is commonly used. After the 1D pooling operation, the output feature The mathematical expression is as follows:

[0096]

[0097] In the formula: is the length of the output feature after 1D pooling, The step size for pooling. The value of the element within the pooling kernel. and These represent the length and width of the pooling kernel, respectively.

[0098] The role of the 1D spatial attention layer is to focus on effective feature information, select and extract effective fault features for subsequent layers, and output the processed features. Represented as:

[0099]

[0100] In the formula: For the first The first channel Each feature element value For the first The first channel Each feature weight coefficient.

[0101] The role of the Softmax output layer is to normalize the output values ​​of the 1D spatial attention layer and convert them into probability outputs. The expression for the Softmax output is:

[0102]

[0103] In the formula, Indicates the first The first channel The exponential value pairs of each feature, where The range is [ 1 , o n ] , This represents the number of fault categories output by the model.

[0104] The AM-CNN model for high-voltage circuit breaker anomaly recognition was trained using vibration data envelopment lines and state labels from a sample database. A schematic diagram of the model training is attached. Figure 6 As shown, firstly, an AM-CNN structure is established, and training parameters are configured. The results are calculated in the order of the input data layer, 1D convolutional layer, 1D pooling layer, 1D spatial attention layer, and Softmax output layer. To make the calculated results close to the actual labels, the model needs continuous training and optimization. Then, the deviation between the actual labels and the forward calculation results is continuously calculated, and the deviation is compared with a set error threshold. If the deviation does not meet the error threshold requirement, the network parameters are optimized using the backpropagation algorithm. If the deviation meets the error threshold requirement, the training is complete, the model parameters are saved, and the trained anomaly recognition model is applied online. In this embodiment, the set error threshold is 0.001.

[0105] Step 3: Collect the operation vibration signal of the high-voltage circuit breaker in actual operation. In actual application, the vibration signal of the high-voltage circuit breaker is collected in real time by using a data collection card. Different from step 1.1, the state of the circuit breaker is not known in advance, and the subsequent high-voltage circuit breaker fault identification model needs to be used for diagnosis.

[0106] The vibration signal of the actual operation is filtered by using wavelet transform to filter out high-frequency interference components. The processing process is referred to step 1.3.

[0107] The sliding root-mean-square envelope of the denoised actual operation vibration signal is extracted based on the Envelope function. The processing process is referred to step 1.4.

[0108] Step 4: The envelope of the denoised actual operation vibration signal obtained in step 3 is input into the AM-CNN-based high-voltage circuit breaker abnormality identification model trained in step 2, and the running state of the high-voltage circuit breaker is output. Then, according to the output running state of the high-voltage circuit breaker, the specific abnormality of the high-voltage circuit breaker is judged: if the output is normal state, the vibration signal envelope collected online is automatically given a normal state label and improved to the sample library; if the output is abnormal state, step 5 is executed. The abnormal state of the high-voltage circuit breaker includes the abnormal state and the unknown state.

[0109] Step 5: Manually check the specific fault state of the high-voltage circuit breaker and improve the sample library.

[0110] If the high-voltage circuit breaker state is judged as abnormal state in step 4, it means that the high-voltage circuit breaker is in abnormal state or unknown state, and the operation and maintenance personnel need to check the actual fault state of the high-voltage circuit breaker and repair it. Among them, the abnormal state includes the control loop abnormal fault state, the spring loosening fault state and the transmission mechanism jamming fault state. If the model automatically identifies the abnormal state of the high-voltage circuit breaker consistent with the maintenance state of the operation and maintenance personnel, the identification consistent instruction is input to the system, the vibration signal envelope collected online is automatically given a running state label and the sample library is improved. If the abnormality identification model automatically identifies the abnormal state of the circuit breaker inconsistent with the maintenance state or appears unknown abnormal state, the operation and maintenance personnel need to manually add the running state label to the vibration signal envelope collected online, and the system updates the sample database.

[0111] After the continuous improvement of the normal state and abnormal state sample library of the high-voltage circuit breaker, the high-voltage circuit breaker fault diagnosis system periodically relearns the newly generated sample library data based on the AM-CNN-based high-voltage circuit breaker abnormality identification model, so that the generalization ability of the AM-CNN-based high-voltage circuit breaker abnormality identification model is continuously enhanced.

[0112] In order to illustrate the effectiveness of the high-voltage circuit breaker abnormality identification method designed by the present application, the present embodiment compares the identification results of the method of the present application with the results of traditional different high-voltage circuit breaker fault diagnosis methods. Figure 7 For the fault diagnosis rates of different fault diagnosis algorithms on the four existing states, in the figure, S1 represents the normal state, S2 represents the control loop abnormality fault state, S3 represents the spring loosening fault state, and S4 represents the transmission mechanism jamming fault state. For states S1 and S3, methods 1, 2, 3 and 4 all have a certain amount of misdiagnosis; and for states S1 and S4, all methods can achieve 100% accurate fault diagnosis, and the accuracy of the fault identification of the four states of the method of the present application all reaches 100%. Figure 8 For the overall accuracy and time consumption of different fault diagnosis methods, it can be seen from the figure that the accuracy of the identification method of the present application is at least 5% higher than that of other methods, and the time consumption is the lowest.

[0113] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A method for identifying anomalies in high-voltage circuit breakers based on vibration signal envelopes, characterized in that, Includes the following steps: Step 1: Based on the conditional generative adversarial network model, generate vibration signal data of several high-voltage circuit breakers corresponding to specified operating states; after filtering the vibration signals, extract the envelope of each vibration signal and associate it with the operating state label to form a sample library; The operating status includes normal status, abnormal control circuit fault status, loose spring fault status, and stuck transmission mechanism fault status. Step 2: Construct a high-voltage circuit breaker anomaly identification model. The model input is the envelope of the vibration signal and the corresponding operating status label, and the model output is the operating status of the high-voltage circuit breaker. The high-voltage circuit breaker anomaly identification model includes an input layer, a 1D convolutional layer, a 1D pooling layer, a 1D spatial attention layer, and a Softmax output layer connected in sequence. The high-voltage circuit breaker anomaly identification model is trained using the sample library data from step 1; Step 3: Obtain the vibration signal of the actual high-voltage circuit breaker, filter and denoise the vibration signal to remove high-frequency interference components, and extract the envelope of the denoised vibration signal; Step 4: Input the envelope of the vibration signal obtained in Step 3 into the trained high-voltage circuit breaker anomaly identification model. The high-voltage circuit breaker anomaly identification model predicts and outputs the operating status of the high-voltage circuit breaker. If the operating status is normal, the envelope of the vibration signal obtained in Step 3 is automatically labeled and the sample library is updated. If the operating status is abnormal, Step 5 is executed. The abnormal state includes both abnormal and unknown states; Step 5: Manually check the abnormal status of the high-voltage circuit breaker and compare the abnormal status determined by manual inspection with the abnormal status output by the high-voltage circuit breaker abnormal identification model. If they are consistent, the envelope of the vibration signal obtained in Step 3 is automatically labeled and the sample library is updated. If they are inconsistent or the status is unknown, the envelope of the vibration signal obtained in Step 3 is manually labeled and the sample library is updated.

2. The high-voltage circuit breaker anomaly identification method according to claim 1, characterized in that, Step 1 includes the following steps: Step 1.1: Perform the closing or opening operation of the high-voltage circuit breaker and collect multiple sets of real vibration signals of the high-voltage circuit breaker under normal and abnormal conditions; among them, the abnormal conditions include abnormal control circuit fault conditions, spring loosening fault conditions, and transmission mechanism jamming fault conditions, which are realized through artificial simulation. Step 1.2: Construct a generative adversarial network model, and learn the distribution characteristics of the real vibration signal data obtained in Step 1.1 through adversarial training between the generator and discriminator in the generative adversarial network model; use the trained generator to synthesize vibration signal data; Step 1.3: Use wavelet transform to filter and reduce noise in the vibration signal data obtained in Steps 1.1 and 1.2; Step 1.4: Use the Envelope function to extract the envelope of the vibration signal after noise reduction and filtering, and create operation status labels for the envelope of the vibration signal according to the corresponding operation status to form a sample library.

3. The high-voltage circuit breaker anomaly identification method according to claim 2, characterized in that, In step 1.1, the corresponding abnormal state is simulated in the following way: Abnormal fault state of the control circuit: Reduce the control voltage of the high-voltage circuit breaker control panel. The above-mentioned spring loosening fault condition: First remove the closing spring and all the shims at the connection of the closing spring in the operating mechanism of the high-voltage circuit breaker, and then reinstall the closing spring back into the operating mechanism. The transmission mechanism jamming fault state: a heavy metal object is suspended at the connection of the transmission mechanism to increase the operating resistance of the transmission mechanism.

4. The high-voltage circuit breaker anomaly identification method according to claim 1, characterized in that, After step 5, the method further includes: training and optimizing the high-voltage circuit breaker anomaly identification model using the updated sample library.

5. A computer-readable storage medium, characterized in that: The device contains a computer-executable program, which, when executed, implements the high-voltage circuit breaker anomaly identification method according to any one of claims 1 to 4.

6. An electronic device for data processing, characterized in that: The device includes one or more processors and a computer-readable storage medium as described in claim 5, for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the high-voltage circuit breaker anomaly identification method as described in any one of claims 1 to 4.

Images