A method and system for evaluating the operating status of high-voltage switches based on intelligent sensing

Abstract

The present invention relates to the technical field of high-voltage switch operating state evaluation, and discloses a method and system for evaluating the operating state of a high-voltage switch based on intelligent sensing, including obtaining vibration waveform data of the high-voltage switch drive mechanism during operation and removing noise; extracting waveform feature parameters from the vibration waveform data according to Fourier transform and wavelet transform, inputting the waveform feature parameters into the mechanism state classification prediction model to obtain the meshing clearance working condition; obtaining the relative amplitude according to the vibration waveform data, obtaining abnormal waveform data according to the comparison relationship between the relative amplitude and the relative amplitude threshold, and extracting abnormal feature parameters from the abnormal waveform data; inputting the abnormal feature parameters into the state scoring model to obtain the state score, and determining the operating state of the high-voltage switch drive mechanism according to the state score. The present invention realizes the comprehensive monitoring and accurate evaluation of the operating state of the high-voltage switch drive mechanism, and ensures the safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage switch operation state evaluation, and particularly to a method and system for evaluating the operation state of a high-voltage switch based on intelligent sensing. Background Art

[0002] The reliable operation of high-voltage switches is crucial for the stability of the power system, and the state of its transmission mechanism directly determines the performance of the switch, thus affecting the stability and reliability of the power system. Traditional condition monitoring methods mainly rely on regular maintenance and manual inspection. This method is not only inefficient but also difficult to detect potential faults in real time. In recent years, condition monitoring technologies based on vibration signal analysis have gradually become a research hotspot.

[0003] In practical engineering applications, due to the long-term action of mechanical stress and electromagnetic force on the transmission system of the transmission mechanism, the meshing clearance of the transmission shaft will change to a certain extent, resulting in a change in the vibration waveform during the opening and closing process of the switch. That is to say, the vibration waveform data can reflect the operation state of the high-voltage switch transmission mechanism, especially the change of the meshing clearance. Therefore, in-depth analysis of the internal relationship between the change of the transmission shaft meshing clearance and the vibration waveform characteristics is of great significance for accurately evaluating the state of the high-voltage switch transmission mechanism. However, since the vibration signal usually contains a large amount of noise, and the accuracy of feature extraction and state classification is limited. In the existing technology, the feature extraction method is single, and the complexity and accuracy of the state classification model are insufficient, making it difficult to achieve precise evaluation of the operation state of the high-voltage switch transmission mechanism. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method and system for evaluating the operation state of a high-voltage switch based on intelligent sensing, so as to solve the problems of low efficiency of the existing method and difficulty in detecting potential faults in real time, and achieve the technical effects of accurately evaluating the operation state of the high-voltage switch transmission mechanism and improving the reliability and safety of the operation of power equipment.

[0005] In the first aspect, the present invention provides a method for evaluating the operation state of a high-voltage switch based on intelligent sensing, and the method includes:

[0006] Obtain the vibration waveform data of the high-voltage switch transmission mechanism during operation, and remove the noise from the vibration waveform data to obtain the denoised vibration waveform data;

[0007] Extract features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, and the waveform feature parameters include frequency band feature parameters and wavelet feature parameters;

[0008] Input the waveform feature parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance condition of the high-voltage switch drive mechanism, where the mechanism state classification prediction model is constructed based on a convolutional neural network and a Markov chain;

[0009] According to the longitudinal component and the transverse component of the vibration waveform data after denoising, obtain the relative amplitude. According to the comparison relationship between the relative amplitude and the relative amplitude threshold, obtain the abnormal waveform data, and extract the abnormal feature parameters from the abnormal waveform data;

[0010] Input the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determine the operating state of the high-voltage switch drive mechanism under the meshing clearance condition according to the state score.

[0011] Further, the step of performing noise removal on the vibration waveform data to obtain the vibration waveform data after denoising includes:

[0012] Use a low-pass filter to filter the data sequence of the vibration waveform data. The data sequence includes a longitudinal data sequence and a transverse data sequence, and the cut-off frequency of the low-pass filter is determined based on the main frequency of the signal in the vibration waveform data;

[0013] Use the spectrum analysis method to extract the vibration amplitude from the filtered data sequence. According to the comparison relationship between the vibration amplitude and the vibration amplitude threshold, perform data segment marking on the filtered data sequence to obtain several marked data segments;

[0014] Perform secondary filtering on each marked data segment based on an adaptive filter, and use a sliding window mechanism to segment the filtered marked data segments;

[0015] Within each sliding window, calculate the vector synthesis value of the longitudinal data and the transverse data to obtain a synthetic vibration waveform curve, and fit the synthetic vibration waveform curve to obtain a fitted waveform curve;

[0016] Correct the vibration waveform data according to the fitted waveform curve to obtain the vibration waveform data after denoising.

[0017] Further, the step of performing feature extraction on the vibration waveform data after denoising according to Fourier transform and wavelet transform to obtain waveform feature parameters includes:

[0018] Use Fourier transform to perform time-frequency conversion on the vibration waveform data after denoising to obtain the frequency-domain amplitude, and extract the frequency-band feature parameters corresponding to each frequency band from the frequency-domain amplitude according to a preset frequency band. The frequency-band feature parameters include the maximum amplitude, the average amplitude, and the frequency bandwidth;

[0019] The vibration waveform data after denoising is decomposed into three layers by wavelet transform to obtain wavelet feature parameters, and the wavelet feature parameters include energy value, standard value, and peak value;

[0020] The frequency band feature parameters and the wavelet feature parameters are normalized to obtain waveform feature parameters.

[0021] Further, the mechanism state classification prediction model includes a convolutional neural network model and a hidden Markov model;

[0022] The convolutional neural network model contains three convolutional layers and two fully connected layers. A pooling layer is connected after each convolutional layer. The size of the convolutional kernel in the convolutional layer decreases with the number of layers. The pooling layer uses the maximum pooling method, and the activation function uses the rectified linear unit. The input of the convolutional neural network model is the waveform feature parameters, and the output is the clearance level probability;

[0023] The number of states of the hidden Markov model is the same as the number of clearance levels. The waveform feature parameters within a continuous time window are used as the observation sequence, and the clearance level probability output by the convolutional neural network model is used as the observation probability matrix. The state transition matrix is calculated from the training data using the maximum likelihood estimation. The output of the hidden Markov model is the state sequence probability calculated based on the state transition matrix and the observation probability matrix;

[0024] The clearance level probability output by the convolutional neural network model and the state sequence probability output by the hidden Markov model are weighted and summed, and the meshing clearance condition determined according to the summed probability is used as the output of the mechanism state classification prediction model.

[0025] Further, the steps of obtaining the relative amplitude according to the longitudinal component and the transverse component of the vibration waveform data after denoising and obtaining the abnormal waveform data according to the comparison relationship between the relative amplitude and the relative amplitude threshold include:

[0026] Based on the sliding window mechanism, the vibration waveform data after denoising is segmented. According to the longitudinal component and the transverse component in each segment of data, the relative amplitude is calculated, and the relative amplitude is filtered to obtain the filtered relative amplitude;

[0027] The filtered relative amplitude is compared with the relative amplitude threshold, and the vibration waveform data exceeding the amplitude threshold is used as the abnormal waveform data. The amplitude threshold is determined based on the mean and variance of the relative amplitude distribution model, and the relative amplitude distribution model is constructed based on the historical vibration waveform data.

[0028] Further, the steps of extracting abnormal feature parameters from the abnormal waveform data include:

[0029] Calculate the abnormal waveform data to obtain time-domain characteristic parameters, where the time-domain characteristic parameters include the time-domain mean, the time-domain variance, and the time-domain peak factor;

[0030] Perform time-frequency conversion on the abnormal waveform data using Fourier transform, and extract frequency-domain characteristic parameters from the converted frequency-domain data, where the frequency-domain characteristic parameters include the main frequency amplitude, the frequency band energy ratio, and the harmonic coefficient;

[0031] Perform three-layer decomposition on the abnormal waveform data using wavelet transform to obtain wavelet characteristic parameters, where the wavelet characteristic parameters include the energy value, the standard value, and the peak value;

[0032] Use the time-domain characteristic parameters, the frequency-domain characteristic parameters, and the wavelet characteristic parameters as abnormal characteristic parameters.

[0033] Further, the step of inputting the abnormal characteristic parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determining the operating state of the high-voltage switch under the meshing clearance condition according to the state score includes:

[0034] Input the abnormal characteristic parameters into a state scoring model constructed based on a deep neural network to obtain a state score;

[0035] Perform Gaussian kernel density estimation on the state score to obtain a probability density function, and calculate the time proportion of the state scores lower than the score threshold based on the probability density function;

[0036] Determine the operating state of the high-voltage switch under the meshing clearance condition according to the time proportion.

[0037] Further, after the step of determining the operating state of the high-voltage switch under the meshing clearance condition, it further includes:

[0038] Extract the historical clearance change amount, the historical vibration waveform data, and the meshing clearance condition from the historical records of the transmission mechanism state assessment;

[0039] Establish a mapping relationship between the amplitude change rate and the clearance change amount according to the historical clearance change amount and the historical vibration waveform data, and obtain the clearance change critical point according to the mapping relationship;

[0040] Extract the waveform characteristic parameters within the time window before and after the clearance change critical point from the historical vibration waveform data, and obtain the parameter change rate according to the extracted waveform characteristic parameters;

[0041] Statistically evaluate the accuracy rate of the meshing clearance condition based on the historical clearance change amount and the meshing clearance condition, establish a probability density function of the accuracy rate distribution according to the statistical results, and extract the mean value of the evaluation accuracy rate and the upper and lower limits of the accuracy rate confidence interval from the probability density function of the accuracy rate distribution;

[0042] Generate a table of the influence on the state evaluation accuracy according to the corresponding relationship between the historical clearance change amount, the parameter change rate, the mean value of the evaluation accuracy rate, and the upper and lower limits of the accuracy rate confidence interval.

[0043] Further, the steps of establishing a mapping relationship between the amplitude change rate and the clearance change amount according to the historical clearance change amount and the historical vibration waveform data, and obtaining the clearance change critical point according to the mapping relationship include:

[0044] Process the historical vibration waveform data using a third-order autoregressive algorithm to obtain a sequence of relative amplitude change rates;

[0045] According to the sequence of relative amplitude change rates and the historical clearance change amount, use a support vector regression algorithm with a radial basis kernel function to establish a mapping relationship between the amplitude change rate and the clearance change amount;

[0046] Determine the clearance change critical point according to the mapping relationship and the slope mutation point of the amplitude change rate curve.

[0047] In a second aspect, the present invention provides a high-voltage switch operating state evaluation system based on intelligent sensing, and the system includes:

[0048] A data preprocessing module for acquiring vibration waveform data of the high-voltage switch driving mechanism during operation and removing noise from the vibration waveform data to obtain denoised vibration waveform data;

[0049] A feature extraction module for extracting waveform feature parameters from the denoised vibration waveform data according to Fourier transform and wavelet transform, and the waveform feature parameters include frequency band feature parameters and wavelet feature parameters;

[0050] A clearance condition prediction module for inputting the waveform feature parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance condition of the high-voltage switch driving mechanism, and the mechanism state classification prediction model is constructed based on a support vector machine model;

[0051] An abnormality determination module for obtaining a relative amplitude according to the longitudinal component and the transverse component of the denoised vibration waveform data, obtaining abnormal waveform data according to the comparison relationship between the relative amplitude and a relative amplitude threshold, and extracting abnormal feature parameters from the abnormal waveform data;

[0052] A state evaluation module is configured to input the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determine the operating state of the high-voltage switch driving mechanism under the meshing clearance condition according to the state score.

[0053] The present invention provides a method and system for evaluating the operating state of a high-voltage switch based on intelligent sensing. By combining Fourier transform, wavelet transform, convolutional neural network, and Markov chain, the present invention can comprehensively capture the characteristics of vibration signals, improve the accuracy of state classification and scoring; through the calculation of relative amplitudes and the rapid extraction of abnormal waveform data, potential faults can be discovered in real time, improving the timeliness of state monitoring; at the same time, by combining a deep neural network and a Markov chain, the complexity and uncertainty of vibration signals can be effectively addressed, improving the robustness of state evaluation. The present invention can effectively improve the state monitoring level of the high-voltage switch driving mechanism, thereby reducing the equipment failure rate and ensuring the safe and stable operation of the power system. Brief Description of the Drawings

[0054] Figure 1 is a schematic flowchart of the method for evaluating the operating state of a high-voltage switch in an embodiment of the present invention;

[0055] Figure 2 is a schematic structural diagram of the system for evaluating the operating state of a high-voltage switch in an embodiment of the present invention. Detailed Embodiments

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Please refer to Figure 1 , a method for evaluating the operating state of a high-voltage switch based on intelligent sensing proposed in the first embodiment of the present invention, which includes steps S10 to S50:

[0058] Step S10: Obtain the vibration waveform data of the high-voltage switch driving mechanism during operation, and remove the noise from the vibration waveform data to obtain the denoised vibration waveform data;

[0059] Step S20: Extract features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, where the waveform feature parameters include frequency band feature parameters and wavelet feature parameters;

[0060] Step S30: Input the waveform feature parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance condition of the high-voltage switch drive mechanism. The mechanism state classification prediction model is constructed based on a support vector machine model.

[0061] Step S40: Obtain the relative amplitude according to the longitudinal and transverse components of the denoised vibration waveform data. Based on the comparison relationship between the relative amplitude and the relative amplitude threshold, obtain the abnormal waveform data, and extract the abnormal feature parameters from the abnormal waveform data.

[0062] Step S50: Input the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determine the operating state of the high-voltage switch drive mechanism under the meshing clearance condition according to the state score.

[0063] The present invention evaluates the operating state of a high-voltage switch by analyzing the vibration waveform data of the drive mechanism of the high-voltage switch. First, vibration data is collected through a vibration sensor array installed at each key node of the drive mechanism. Among them, the sensor array is arranged orthogonally in the longitudinal and transverse directions, the sampling frequency is set as the number of samples per second, and the sampling time is the whole process of the drive mechanism's action. Since the vibration signal of the drive mechanism of the high-voltage switch has typical bidirectional component characteristics, the orthogonal arrangement of the vibration sensors can better capture the propagation law of the vibration wave. Since the original vibration waveform data collected by the sensors is subject to noise interference, it is also necessary to remove the noise from the collected vibration waveform data. The denoising method can use a conventional filter to filter out the noise, or use a method of correcting the vibration waveform data based on a calibrated curve obtained through testing to remove the noise.

[0064] In a preferred embodiment, to improve the denoising effect, the present invention provides a method for denoising vibration waveform data based on a fitting curve. The specific steps include:

[0065] Filter the data sequence of the vibration waveform data using a low-pass filter. The data sequence includes a longitudinal data sequence and a transverse data sequence. The cut-off frequency of the low-pass filter is determined based on the main frequency of the signal in the vibration waveform data.

[0066] Extract the vibration amplitude from the filtered data sequence using a spectrum analysis method. Based on the comparison relationship between the vibration amplitude and the vibration amplitude threshold, mark the data segments of the filtered data sequence to obtain several marked data segments.

[0067] Perform secondary filtering on each marked data segment based on an adaptive filter, and segment the filtered marked data segments using a sliding window mechanism.

[0068] Within each sliding window, calculate the vector synthesis value of the longitudinal data and the transverse data to obtain a synthesized vibration waveform curve, and fit the synthesized vibration waveform curve to obtain a fitted waveform curve;

[0069] Correct the vibration waveform data according to the fitted waveform curve to obtain the vibration waveform data after denoising.

[0070] In this embodiment, a longitudinal data sequence and a transverse data sequence are divided from the vibration waveform data, and the two-direction data sequences are respectively processed by a Butterworth low-pass filter. Among them, the cut-off frequency of the filter is half of the main frequency of the signal, and the main frequency of the signal is obtained by calculating the spectrum of the data sequence through Fourier transform and acquiring it from the spectrum.

[0071] For the filtered longitudinal data sequence and transverse data sequence, the vibration amplitude is extracted from the data sequence by using the spectrum analysis method. The vibration amplitude is also the absolute amplitude, which is used to describe the intensity or energy of the vibration and can be characterized by the instantaneous amplitude, peak value or root mean square value. When the vibration amplitude exceeds the vibration amplitude threshold, the data segment is marked. Among them, the vibration amplitude threshold is the upper limit of the mean value of the vibration amplitude in the data collected by the sensor at the same position in the historical operation record.

[0072] According to the start and end times of the marked data segment, an adaptive filter is constructed for this data segment. The order of the filter is set to be equal to the length of the data segment, and the recursive least squares method is used to update the filter coefficients. The data segment is subjected to secondary filtering processing based on the updated filter coefficients. Then, the sliding window method is used to segment the data segment after filtering processing. The window length is set to an integer multiple of the data sampling period, and the overlapping rate of adjacent windows is set to 50%. Within each sliding window, calculate the vector synthesis value of the longitudinal and transverse vibration data to obtain a synthesized vibration waveform curve. Support vector regression is used to fit the synthesized vibration waveform curve within the window, and the abnormal points in the original vibration waveform data that deviate from the fitted curve are corrected according to the fitted curve, and the correction amplitude is limited to not exceed 20% of the original data to obtain the finally denoised vibration waveform data.

[0073] Through the denoising method provided in this embodiment, not only can various abnormal fluctuations be removed, but also the true characteristics of the vibration signal can be retained to a large extent, thus providing accurate data support for subsequent waveform data analysis and processing.

[0074] For the denoised vibration waveform data, feature extraction is performed through Fourier transform and wavelet transform. The specific steps include:

[0075] The Fourier transform is used to perform time-frequency conversion on the denoised vibration waveform data to obtain the frequency-domain amplitude, and according to a preset frequency band, the band feature parameters corresponding to each frequency band are extracted from the frequency-domain amplitude. The band feature parameters include the maximum amplitude, the average amplitude, and the frequency bandwidth.

[0076] The wavelet transform is used to perform three-layer decomposition on the denoised vibration waveform data to obtain wavelet feature parameters. The wavelet feature parameters include the energy value, the standard value, and the peak value.

[0077] The band feature parameters and the wavelet feature parameters are normalized to obtain waveform feature parameters.

[0078] In this embodiment, first, the Fourier transform is used to convert the vibration waveform data into frequency-domain data, thereby obtaining the frequency-domain features, that is, the amplitude spectrum. According to the frequency-domain features, the frequency-domain amplitude is obtained. At the same time, according to the frequency-domain features, the vibration frequencies are divided into three frequency bands: the low-frequency band, the medium-frequency band, and the high-frequency band according to the vibration frequency range, and the corresponding band feature parameters are extracted according to the frequency band. The band feature parameters include the maximum amplitude, the average amplitude, and the frequency bandwidth.

[0079] Then, the wavelet transform is used to perform wavelet transform on the time-domain sequences of the longitudinal component and the transverse component respectively, and three-layer decomposition is performed to obtain four coefficient arrays of high frequency, medium frequency, low frequency, and the lowest frequency approximation. For each layer of decomposition, the energy value, the standard value, and the peak value are extracted as the wavelet feature parameters. Among them, the energy value is the sum of the squares of the coefficient array, the standard deviation is the standard deviation of the coefficient array, and the peak value is the absolute maximum value of the coefficient array. Finally, the maximum-minimum linear transformation is used to normalize the above band feature parameters and wavelet feature parameters to obtain a normalized vector, that is, the waveform feature parameters are obtained.

[0080] The waveform feature parameters will be input into a pre-trained mechanism state classification prediction model for mechanism state prediction. The mechanism state classification prediction model is used to predict the meshing clearance condition of the high-voltage switch drive mechanism. In practical applications, the meshing clearance state of the transmission shaft of the high-voltage switch drive mechanism is closely related to the vibration characteristics of the drive mechanism. Therefore, by analyzing the vibration characteristics, different clearance conditions can be effectively identified.

[0081] In this embodiment, the mechanism state classification prediction model can be constructed based on the support vector machine model. The radial basis kernel function is selected as the kernel function, and the grid search method is used to optimize the kernel function parameters and the penalty factor. The trained classification prediction model is used to calculate the distance from the vibration feature parameters to each category hyperplane, and the probability value belonging to each type is calculated according to the distance value, and the category corresponding to the maximum probability is selected as the prediction result, so as to obtain the meshing clearance condition of the high-voltage switch drive mechanism under the current vibration feature parameters.

[0082] In a preferred embodiment, the institutional state classification prediction model can also be constructed based on a convolutional neural network and a Markov chain. In this embodiment, the institutional state classification prediction model includes a convolutional neural network model and a hidden Markov model. Among them, the convolutional neural network model contains three convolutional layers, and a pooling layer is connected behind each convolutional layer. The size of the convolutional kernel of the convolutional layer decreases with the number of layers. The pooling layer adopts the maximum pooling method, and the activation function adopts the rectified linear unit ReLU. Two fully connected layers are connected behind the pooling layer. The first fully connected layer flattens the output of the convolutional layer and connects it to a hidden layer, and the second fully connected layer is connected to the output layer. The number of neurons is equal to the number of clearance levels, and the output layer uses the Softmax function to output the probability of each clearance level.

[0083] When training the convolutional neural network model, according to the preset meshing clearance range of the transmission shaft, it is evenly divided into five clearance levels from the minimum clearance value to the maximum clearance value. Under each clearance level, the vibration response of the transmission shaft is calculated by finite element simulation. The simulation boundary conditions include the constraint positions and material parameters at both ends of the transmission shaft. Vibration waveform data is extracted from the vibration response, and a training dataset is constructed through data processing to train the convolutional neural network model. After the training is completed, the obtained convolutional neural network model can be used as a clearance level predictor to predict the clearance level probability distribution of the input feature vector sequence.

[0084] To ensure the accuracy of the prediction, a hidden Markov model is added to the classification prediction model. When constructing the hidden Markov model, the number of its states is defined to be the same as the number of clearance levels, and each state corresponds to a clearance level. The observation sequence is the feature vector sequence within a continuous time window. The initial state probability distribution is defined as a uniform distribution. Based on the maximum likelihood estimation, the state transition matrix is calculated from the training data, and the clearance level probability predicted by the convolutional neural network model is used as the observation probability matrix. Based on the above definitions, the hidden Markov model performs forward probability calculation and backward probability calculation according to the state transition matrix and the observation probability matrix, and obtains the state sequence probability according to the forward probability and the backward probability.

[0085] Finally, the gap level probability output by the convolutional neural network model and the state sequence probability output by the hidden Markov model are weighted and summed to obtain the gap condition probability. Based on the gap condition probability, the current meshing gap condition is determined, and the meshing gap condition is used as the output of the classification prediction model. The mechanism state classification prediction model provided by this embodiment can accurately predict the meshing gap condition of the high-voltage switch drive mechanism. It should be noted that the above embodiment is only a preferred model construction method. The mechanism state classification prediction model in the present invention can also be constructed only based on the convolutional neural network model, or constructed using other neural network models such as deep neural network models, or constructed using machine learning algorithms, and no more limitations are made here.

[0086] In the present invention, for the evaluation of the operating state of the high-voltage switch, in addition to evaluating the current meshing gap condition, it is also necessary to evaluate whether the operating state of the drive mechanism is normal or abnormal under the current meshing gap condition. First, based on the comparison relationship between the relative amplitude and the relative amplitude threshold, abnormal waveform data is extracted from the vibration waveform data. The specific steps include:

[0087] Based on the sliding window mechanism, the denoised vibration waveform data is segmented. According to the longitudinal component and the transverse component in each segment of data, the relative amplitude is calculated, and the relative amplitude is filtered to obtain the filtered relative amplitude.

[0088] The filtered relative amplitude is compared with the relative amplitude threshold, and the vibration waveform data exceeding the amplitude threshold is used as abnormal waveform data. The amplitude threshold is determined based on the mean and variance of the relative amplitude distribution model, and the relative amplitude distribution model is constructed based on historical vibration waveform data.

[0089] In this embodiment, first, the denoised vibration waveform data is segmented using the sliding window mechanism. By cutting the continuous vibration data into multiple smaller data segments, it is convenient for subsequent analysis. Preferably, the fixed length of the sliding window is set to one-fiftieth of the number of sampling points, and the sliding step is set to half of the window length. For each data segment, the longitudinal component and the transverse component are extracted respectively. In vibration analysis, the longitudinal component refers to the displacement or acceleration component consistent with the vibration direction, while the transverse component refers to the component perpendicular to the vibration direction. Then, the relative amplitude of the longitudinal component and the transverse component in each segment of data is calculated. The relative amplitude can be obtained by comparing the absolute values, differences, or root mean square values of the two components. In addition, a Kalman filter is used to smooth the relative amplitude, and data smoothing is achieved by reasonably setting the noise covariance.

[0090] Then, compare the filtered relative amplitude with the relative amplitude threshold, and regard the vibration waveform data corresponding to the relative amplitude exceeding the amplitude threshold as abnormal waveform data. In this embodiment, a normal distribution model of the relative amplitude is constructed according to the historical operation records, the mean and standard deviation of the relative amplitude in the normal operation state are calculated, and the mean plus or minus three times the standard deviation is used as the preset relative amplitude threshold range.

[0091] After the abnormal waveform data is extracted, analyze the abnormal waveform data to determine whether there is an abnormality in the operation state of the transmission mechanism. In this embodiment, the analysis steps include two parts: extraction of abnormal characteristic parameters and scoring of the operation state. Among them, the specific steps for extracting abnormal characteristic parameters include:

[0092] Calculate the abnormal waveform data to obtain time-domain characteristic parameters, where the time-domain characteristic parameters include time-domain mean, time-domain variance, and time-domain peak factor;

[0093] Perform time-frequency conversion on the abnormal waveform data using Fourier transform, and extract frequency-domain characteristic parameters from the converted frequency-domain data. The frequency-domain characteristic parameters include main frequency amplitude, frequency band energy ratio, and harmonic coefficient;

[0094] Perform three-layer decomposition on the abnormal waveform data using wavelet transform to obtain wavelet characteristic parameters, where the wavelet characteristic parameters include energy value, standard value, and peak value;

[0095] Regard the time-domain characteristic parameters, the frequency-domain characteristic parameters, and the wavelet characteristic parameters as abnormal characteristic parameters.

[0096] In this embodiment, three types of parameters are used as abnormal characteristic parameters, namely time-domain characteristic parameters, frequency-domain characteristic parameters, and wavelet characteristic parameters. Among them, the time-domain characteristic parameters include three types: mean, variance, and peak factor. These three types of time-domain characteristic parameters can be directly calculated from the abnormal waveform data; the frequency-domain characteristic parameters include three types: main frequency amplitude, frequency band energy ratio, and harmonic coefficient. The extraction of these three types of frequency-domain characteristic parameters requires first converting the abnormal waveform data from the time domain to the frequency domain through Fourier transform, and then finding the frequency point with the largest amplitude in the spectrum. Its amplitude is the main frequency amplitude. Divide the spectrum into several frequency bands, such as low frequency, medium frequency, and high frequency. By calculating the energy of each frequency band, determine the frequency band energy ratio of each frequency band to the total energy, and then analyze the spectrum to extract the amplitudes of the fundamental frequency and its harmonic frequencies, and use the ratio of the harmonic amplitude to the fundamental wave amplitude as the harmonic coefficient; the wavelet characteristic parameters include three types: energy value, standard value, and peak value. These three types of wavelet characteristic parameters are the same as the wavelet characteristic parameters in the extraction of waveform characteristic parameters. The extraction steps can refer to the above embodiments and will not be elaborated here one by one.

[0097] For the abnormal feature parameters obtained in the above steps, a state scoring model constructed based on a deep neural network is used to perform state scoring, and the operating state of the high-voltage switch is determined to be normal or abnormal according to the evaluation score. Among them, the state scoring model can be constructed using a deep neural network model or other neural network models, and the operating state of the high-voltage switch can be directly output through the state scoring model.

[0098] In a preferred embodiment, in order to make the evaluation result of the operating state more accurate, the present invention also provides a method for determining the operating state based on the time ratio, and the specific steps include:

[0099] Input the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score;

[0100] Perform Gaussian kernel density estimation on the state score to obtain a probability density function, and based on the probability density function, calculate the time ratio of the state scores lower than the scoring threshold;

[0101] Determine the operating state of the high-voltage switch under the meshing clearance condition according to the time ratio.

[0102] In this embodiment, first, a state scoring model is constructed based on a deep neural network model. Preferably, a deep neural network model with a five-layer structure is used for construction. The number of nodes in each layer decreases sequentially. The ReLU activation function is used in the hidden layer, and the linear activation function is used in the output layer. The score is output based on the weighted sum of each feature score. Then, Gaussian kernel density estimation is performed on the state evaluation score to calculate the probability density function. Assuming that the scoring threshold is 80 points, through the probability density function, all score data are traversed, the number of data points with scores lower than the set threshold is counted, and the time ratio of the state scores lower than the scoring threshold is obtained through the ratio of the number of data points lower than the set threshold to the total number of data points. Finally, it is judged which category of the preset state thresholds this time ratio belongs to, so as to obtain the operating state of the high-voltage switch under the current meshing clearance condition. Preferably, the state thresholds are set as: a normal state with a time ratio less than 5%, a mild abnormality with a time ratio of 5% to 15%, a moderate abnormality with a time ratio of 15% to 30%, and a severe abnormality with a time ratio greater than 30%. Assuming that the calculation result shows that the time ratio of scores lower than the 80-point threshold reaches 30%, based on the state threshold, it is considered that the high-voltage switch is in a moderate abnormal state. Further, in-depth analysis can also be performed based on the probability density function. Assuming that it is found that the longitudinal vibration amplitude increases abnormally while the lateral vibration is basically normal under this condition, it indicates that the transmission mechanism may have asymmetric wear.

[0103] This method for determining the operating state based on a deep neural network and probability density distribution can provide a more accurate and detailed classification of the degree of abnormality compared to using only a neural network model for state evaluation. It reduces the bias that may be brought about by a single model and enhances the robustness and reliability of the evaluation. In addition, more information about the characteristics of vibration data can be obtained from the score distribution, which helps to further analyze the causes and types of faults and provides strong support for subsequent maintenance and management.

[0104] In a preferred embodiment, the present invention also provides a method for analyzing the impact on the accuracy of the evaluation method. The specific steps include:

[0105] Extract the historical clearance change amount, historical vibration waveform data, and meshing clearance working conditions from the historical records of the transmission mechanism state evaluation;

[0106] According to the historical clearance change amount and historical vibration waveform data, establish a mapping relationship between the amplitude change rate and the clearance change amount, and based on this mapping relationship, obtain the clearance change critical point;

[0107] Extract the waveform feature parameters within the time window before and after the clearance change critical point from the historical vibration waveform data, and based on the extracted waveform feature parameters, obtain the parameter change rate;

[0108] According to the historical clearance change amount and the meshing clearance working conditions, statistically analyze the evaluation accuracy rate of the meshing clearance working conditions, establish a probability density function of the accuracy rate distribution based on the statistical results, and extract the average evaluation accuracy rate and the upper and lower limits of the accuracy rate confidence interval from the probability density function of the accuracy rate distribution;

[0109] Generate a state evaluation accuracy impact table based on the corresponding relationship between the historical clearance change amount, the parameter change rate, the average evaluation accuracy rate, and the upper and lower limits of the accuracy rate confidence interval.

[0110] In this embodiment, by analyzing the recorded data in the historical records of the transmission mechanism state evaluation, the relationship between the clearance change amount, the waveform feature parameters, and the accuracy of the evaluation state is determined. Specifically, first, according to the historical clearance change amount and historical vibration waveform data in the historical records, a mapping relationship between the amplitude change rate and the clearance change amount is established to determine the clearance change critical point. The specific steps include:

[0111] Process the historical vibration waveform data using a third-order autoregressive algorithm to obtain a relative amplitude change rate sequence;

[0112] According to the relative amplitude change rate sequence and the historical clearance change amount, use a support vector regression algorithm with a radial basis kernel function to establish a mapping relationship between the amplitude change rate and the clearance change amount;

[0113] Determine the critical point of clearance change according to the mapping relationship and the slope mutation point of the amplitude change rate curve.

[0114] In this embodiment, first, extract the clearance change amount and vibration waveform data from the mechanism state evaluation record, construct an amplitude change trend curve using the third-order autoregressive algorithm, calculate the relative amplitude difference between the longitudinal component and the transverse component of the vibration waveform according to the sampling period, and obtain the change rate sequence by dividing the difference between adjacent sampling points by the time interval.

[0115] Then, according to the relative amplitude change rate sequence of the longitudinal component and the transverse component of the vibration waveform, establish a mapping relationship of the clearance change amount using the support vector regression algorithm with a radial basis kernel function, and determine the critical point of clearance change through the slope mutation point of the amplitude change rate curve, such as the point where the slope is greater than the slope threshold. The change of the meshing clearance will be shown in the change of the vibration characteristic parameters. Therefore, the waveform characteristic parameters within the time window before and after the critical point of clearance change can be extracted, and the parameter change rate can be calculated according to the change of the waveform characteristic parameters before and after.

[0116] Then, perform random sampling through the Monte Carlo method, count the evaluation accuracy rate under each clearance change amount, calculate the mean and standard deviation of the evaluation accuracy rate, establish a normal distribution probability density function based on the accuracy rate statistical results, calculate the accuracy rate probability distribution corresponding to different clearance change amounts, and extract the mean value of the evaluation accuracy rate, the upper and lower limits of the accuracy rate confidence interval from the distribution function. Finally, generate a state evaluation accuracy influence table according to the corresponding relationship between the historical clearance change amount, the parameter change rate, the mean value of the evaluation accuracy rate, and the upper and lower limits of the accuracy rate confidence interval. For example, as the clearance change amount increases, the accuracy rate gradually decreases, and this decreasing trend of the accuracy rate shows an obvious correlation with the vibration amplitude change rate. Through the influence table provided in this embodiment, the change of the waveform characteristic parameters and the change of the state evaluation accuracy rate under different clearance change amounts can be accurately displayed, thus providing an important basis for subsequent equipment maintenance decisions.

[0117] A method for evaluating the operating state of a high-voltage switch based on intelligent sensing provided in this embodiment. The present invention combines Fourier transform, wavelet transform, convolutional neural network, and Markov chain, which can comprehensively capture the characteristics of vibration signals and improve the accuracy of state classification and scoring; through the calculation of relative amplitude and the rapid extraction of abnormal waveform data, potential faults can be discovered in real time, improving the timeliness of state monitoring; at the same time, through the combination of deep neural network and Markov chain, the complexity and uncertainty of vibration signals can be effectively dealt with, improving the robustness of state evaluation. The present invention realizes the comprehensive monitoring and accurate evaluation of the operating state of the high-voltage switch drive mechanism through noise removal, feature extraction, state classification, anomaly detection, and state scoring, effectively improving the state monitoring level of the high-voltage switch drive mechanism, further reducing the equipment failure rate, and ensuring the safe and stable operation of the power system.

[0118] Please refer to Figure 2 , based on the same inventive concept, a system for evaluating the operating state of a high-voltage switch based on intelligent sensing proposed in the second embodiment of the present invention includes:

[0119] A data preprocessing module 10, configured to obtain vibration waveform data of the high-voltage switch drive mechanism during operation, and perform noise removal on the vibration waveform data to obtain denoised vibration waveform data;

[0120] A feature extraction module 20, configured to extract features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, where the waveform feature parameters include frequency band feature parameters and wavelet feature parameters;

[0121] A clearance condition prediction module 30, configured to input the waveform feature parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance condition of the high-voltage switch drive mechanism, where the mechanism state classification prediction model is constructed based on a support vector machine model;

[0122] An anomaly determination module, configured to obtain a relative amplitude according to the longitudinal component and the transverse component of the denoised vibration waveform data, obtain abnormal waveform data according to the comparison relationship between the relative amplitude and a relative amplitude threshold, and extract abnormal feature parameters from the abnormal waveform data;

[0123] A state evaluation module 40, configured to input the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determine the operating state of the high-voltage switch drive mechanism under the meshing clearance condition according to the state score.

[0124] The technical features and technical effects of the high-voltage switch operating state evaluation system based on intelligent sensing proposed in the embodiments of the present invention are the same as those of the method proposed in the embodiments of the present invention, and will not be elaborated here. Each module in the above high-voltage switch operating state evaluation system based on intelligent sensing can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0125] In summary, the embodiments of the present invention propose a method and system for evaluating the operating state of a high-voltage switch based on intelligent sensing. The method includes obtaining vibration waveform data of the high-voltage switch drive mechanism during operation, removing noise from the vibration waveform data to obtain denoised vibration waveform data; extracting features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, where the waveform feature parameters include frequency band feature parameters and wavelet feature parameters; inputting the waveform feature parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance condition of the high-voltage switch drive mechanism, and the mechanism state classification prediction model is constructed based on a convolutional neural network and a Markov chain; obtaining a relative amplitude according to the longitudinal component and the transverse component of the denoised vibration waveform data, obtaining abnormal waveform data according to the comparison relationship between the relative amplitude and the relative amplitude threshold, and extracting abnormal feature parameters from the abnormal waveform data; inputting the abnormal feature parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determining the operating state of the high-voltage switch drive mechanism under the meshing clearance condition according to the state score. The present invention combines Fourier transform, wavelet transform, convolutional neural network, and Markov chain, can comprehensively capture the characteristics of vibration signals, and improve the accuracy of state classification and scoring; through the calculation of relative amplitude and the rapid extraction of abnormal waveform data, potential faults can be found in real time, improving the timeliness of state monitoring; at the same time, through the combination of a deep neural network and a Markov chain, the complexity and uncertainty of vibration signals can be effectively addressed, improving the robustness of state evaluation. The present invention realizes the comprehensive monitoring and accurate evaluation of the operating state of the high-voltage switch drive mechanism through noise removal, feature extraction, state classification, abnormal detection, and state scoring, effectively improving the state monitoring level of the high-voltage switch drive mechanism, thereby reducing the equipment failure rate and ensuring the safe and stable operation of the power system.

[0126] Each embodiment in this specification is described in a progressive manner. For the parts that are the same or similar in each embodiment, reference can be made to each other, and the key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and reference can be made to the corresponding part of the method embodiment for the relevant content. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0127] The above embodiments only represent several preferred embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the technical principle of the present invention, several improvements and substitutions can still be made, and these improvements and substitutions should also be regarded as within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the protection scope of the claims.

Claims

1. A method for evaluating the operating status of a high-voltage switch based on intelligent sensing, characterized in that: include: Acquire vibration waveform data of the high-voltage switch transmission mechanism during operation, and remove noise from the vibration waveform data to obtain denoised vibration waveform data; Perform feature extraction on the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, wherein the waveform feature parameters include frequency band feature parameters and wavelet feature parameters; Inputting the waveform characteristic parameters into a pre-constructed mechanism state classification prediction model to obtain the meshing clearance working condition of the high-voltage switch transmission mechanism, wherein the mechanism state classification prediction model is constructed based on a convolutional neural network and a Markov chain; According to the longitudinal component and the transverse component of the denoised vibration waveform data, a relative amplitude is obtained, according to the comparison relationship between the relative amplitude and the relative amplitude threshold, abnormal waveform data is obtained, and abnormal characteristic parameters are extracted from the abnormal waveform data; Inputting the abnormal characteristic parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determining the operating state of the high-voltage switch transmission mechanism under the meshing clearance working condition according to the state score, including: Performing Gaussian kernel density estimation on the state score to obtain a probability density function, and calculating the time proportion of the state score below the score threshold based on the probability density function; According to the time proportion, the operating state of the high-voltage switch under the meshing gap condition is determined.

2. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1 is characterized in that: The step of removing noise from the vibration waveform data to obtain the denoised vibration waveform data comprises: A low-pass filter is used to filter a data sequence of the vibration waveform data, wherein the data sequence includes a longitudinal data sequence and a transverse data sequence, and a cutoff frequency of the low-pass filter is determined based on a main frequency of a signal in the vibration waveform data; The vibration amplitude is extracted from the filtered data sequence by using a spectrum analysis method, and the filtered data sequence is marked with data segments according to the comparison relationship between the vibration amplitude and the vibration amplitude threshold, so as to obtain a plurality of marked data segments; Perform secondary filtering on each marked data segment based on an adaptive filter, and segment the filtered marked data segment using a sliding window mechanism; In each sliding window, the vector synthesis value of the longitudinal data and the transverse data is calculated to obtain a synthetic vibration waveform curve, and the synthetic vibration waveform curve is fitted to obtain a fitting waveform curve; The vibration waveform data is corrected according to the fitting waveform curve to obtain the denoised vibration waveform data.

3. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1 is characterized in that: The step of extracting features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters comprises: The denoised vibration waveform data is converted into a time-frequency value by Fourier transform to obtain a frequency domain amplitude, and the frequency band characteristic parameters corresponding to each frequency band are extracted from the frequency domain amplitude according to the preset frequency band, wherein the frequency band characteristic parameters include the maximum amplitude, the average amplitude and the frequency bandwidth; The denoised vibration waveform data is decomposed into three layers using wavelet transform to obtain wavelet characteristic parameters, wherein the wavelet characteristic parameters include energy value, standard value and peak value; The frequency band characteristic parameters and the wavelet characteristic parameters are normalized to obtain waveform characteristic parameters.

4. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1 is characterized in that: The institution status classification prediction model includes a convolutional neural network model and a hidden Markov model; The convolutional neural network model comprises three convolutional layers and two fully connected layers. A pooling layer is connected behind each convolutional layer. The convolution kernel size of the convolutional layer decreases with the number of layers. The pooling layer adopts the maximum pooling method, and the activation function adopts the rectified linear unit. The input of the convolutional neural network model is the waveform feature parameter, and the output is the gap level probability. The number of states of the hidden Markov model is the same as the number of gap levels, the waveform characteristic parameters in the continuous time window are used as the observation sequence, the gap level probability output by the convolutional neural network model is used as the observation probability matrix, the state transfer matrix is ​​calculated from the training data using maximum likelihood estimation, and the output of the hidden Markov model is the state sequence probability calculated based on the state transfer matrix and the observation probability matrix; The gap level probability output by the convolutional neural network model and the state sequence probability output by the hidden Markov model are weightedly summed, and the meshing gap working condition determined according to the summed probability is used as the output of the mechanism state classification prediction model.

5. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1, characterized in that: The step of obtaining the relative amplitude according to the longitudinal component and the transverse component of the denoised vibration waveform data, and obtaining the abnormal waveform data according to the comparison relationship between the relative amplitude and the relative amplitude threshold comprises: Based on the sliding window mechanism, the denoised vibration waveform data is segmented, the relative amplitude is calculated according to the longitudinal component and the transverse component in each segment of the data, and the relative amplitude is filtered to obtain the filtered relative amplitude; The filtered relative amplitude is compared with a relative amplitude threshold, and the vibration waveform data exceeding the amplitude threshold is taken as abnormal waveform data. The amplitude threshold is determined based on the mean and variance of a relative amplitude distribution model, and the relative amplitude distribution model is constructed based on historical vibration waveform data.

6. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1, characterized in that: The step of extracting abnormal characteristic parameters from abnormal waveform data comprises: Calculating the abnormal waveform data to obtain time domain characteristic parameters, wherein the time domain characteristic parameters include a time domain mean, a time domain variance, and a time domain peak factor; Using Fourier transform to perform time-frequency conversion on abnormal waveform data, and extracting frequency domain characteristic parameters from the converted frequency domain data, wherein the frequency domain characteristic parameters include main frequency amplitude, frequency band energy ratio and harmonic coefficient; Using wavelet transform to perform three-layer decomposition on abnormal waveform data to obtain wavelet characteristic parameters, wherein the wavelet characteristic parameters include energy value, standard value and peak value; The time domain characteristic parameter, the frequency domain characteristic parameter and the wavelet characteristic parameter are used as abnormal characteristic parameters.

7. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 1, characterized in that: After the step of determining the operating state of the high-voltage switch under the meshing clearance condition, the method further includes: Extract historical clearance changes, historical vibration waveform data and meshing clearance conditions from the transmission mechanism status evaluation history records; According to the historical gap change amount and the historical vibration waveform data, a mapping relationship between the amplitude change rate and the gap change amount is established, and according to the mapping relationship, a gap change critical point is obtained; Extracting waveform characteristic parameters in a time window before and after a critical point of gap change from historical vibration waveform data, and obtaining a parameter change rate based on the extracted waveform characteristic parameters; According to the historical clearance changes and meshing clearance conditions, the evaluation accuracy of the meshing clearance conditions is statistically analyzed, and the accuracy distribution probability density function is established based on the statistical results. The evaluation accuracy mean and the upper and lower limits of the accuracy confidence interval are extracted from the accuracy distribution probability density function. A state assessment accuracy impact table is generated according to the correspondence between the historical gap change amount, the parameter change rate, the assessment accuracy mean value, and the upper and lower limits of the accuracy confidence interval.

8. The method for evaluating the operating status of a high-voltage switch based on intelligent sensing according to claim 7, characterized in that: The step of establishing a mapping relationship between the amplitude change rate and the gap change amount according to the historical gap change amount and the historical vibration waveform data, and obtaining the gap change critical point according to the mapping relationship comprises: The historical vibration waveform data is processed using a third-order autoregressive algorithm to obtain a relative amplitude change rate sequence; According to the relative amplitude change rate sequence and the historical gap change amount, a support vector regression algorithm of a radial basis kernel function is used to establish a mapping relationship between the amplitude change rate and the gap change amount; The gap change critical point is determined according to the mapping relationship and the slope mutation point of the amplitude change rate curve.

9. A high-voltage switch operating status assessment system based on intelligent sensing, characterized in that: include: A data preprocessing module is used to obtain the vibration waveform data of the high-voltage switch transmission mechanism during operation, and remove noise from the vibration waveform data to obtain the denoised vibration waveform data; A feature extraction module, used to extract features from the denoised vibration waveform data according to Fourier transform and wavelet transform to obtain waveform feature parameters, wherein the waveform feature parameters include frequency band feature parameters and wavelet feature parameters; A clearance working condition prediction module, used for inputting the waveform characteristic parameters into a pre-built mechanism state classification prediction model to obtain the meshing clearance working condition of the high-voltage switch transmission mechanism, wherein the mechanism state classification prediction model is constructed based on a support vector machine model; An abnormality determination module is used to obtain a relative amplitude according to the longitudinal component and the transverse component of the denoised vibration waveform data, obtain abnormal waveform data according to a comparison relationship between the relative amplitude and a relative amplitude threshold, and extract abnormal characteristic parameters from the abnormal waveform data; A state evaluation module is used to input the abnormal characteristic parameters into a state scoring model constructed based on a deep neural network to obtain a state score, and determine the operating state of the high-voltage switch transmission mechanism under the meshing clearance working condition according to the state score, including: Performing Gaussian kernel density estimation on the state score to obtain a probability density function, and calculating the time proportion of the state score below the score threshold based on the probability density function; According to the time proportion, the operating state of the high-voltage switch under the meshing gap condition is determined.

Images