Method and system for secondary ac circuit insulation monitoring based on multi-modal machine learning

Abstract

The application belongs to the technical field of circuit automation detection, and specifically provides a secondary alternating current loop insulation monitoring method and system based on multi-modal machine learning, comprising: processing high-frequency pulse current signals generated by insulation defect faults of a secondary loop through multi-element variational modal decomposition to obtain time-frequency domain features of fault signals under different modes; selecting feature parameters for fault diagnosis, using a local linear embedding algorithm to reduce dimensions of feature parameters of fault time domain reconstruction signals and fault frequency domain reconstruction signals, taking the reduced historical fault time-frequency feature vectors and real-time fault time-frequency feature vectors as a training set and a test set respectively, and classifying through a support vector machine algorithm to obtain fault types and fault rates of each fault type after classification. The method and system utilize that insulation defects of a secondary loop will cause partial discharge, and the partial discharge will generate high-frequency pulse current signals, and analyze the insulation state of the secondary loop by extracting features of fault signals.

Description

Technical Field

[0001] This invention belongs to the field of circuit automation testing technology, specifically, it relates to a method and system for monitoring the insulation of secondary AC circuits based on multimodal machine learning. Background Technology

[0002] The secondary circuit of a power system is used to control, protect, regulate, and monitor the operating status of various components in the primary circuit, playing a crucial role in the safe and stable operation of the power system. However, due to the complex wiring and wide scope of the secondary circuit, its defects are often well-hidden. Some defects are only discovered when the fault becomes severe, seriously affecting the correct operation of protection devices and even causing major power grid safety accidents.

[0003] Currently, the main methods for monitoring the insulation status of secondary circuits are to use infrared thermometers to monitor the temperature of the secondary circuit terminal blocks and determine whether the terminals are loose or detached; and to use current transformers to monitor changes in zero-sequence current to determine whether a break-in fault has occurred in the secondary circuit.

[0004] CN106199340A discloses a method and apparatus for detecting open circuits in a CT secondary circuit. This method involves connecting the A, B, and C phase currents in parallel and then using an external zero-sequence CT to collect the zero-sequence current in real time. Simultaneously, the three-phase currents (A, B, and C) and the external zero-sequence current are all connected to an AD sampling channel to calculate the zero-sequence current in real time. The difference between the collected and calculated zero-sequence currents is compared to determine whether the CT secondary circuit is open. However, this method can only determine whether an open circuit fault has occurred in the secondary circuit; it cannot reflect the type and severity of insulation defects such as aging or damage to the insulation layer, or loose or detached terminals.

[0005] CN115436865A discloses a monitoring system for the secondary circuit of a current transformer based on infrared thermal imaging. This system and method address issues such as unplanned power outages caused by high-voltage discharge during open-circuit operation of the secondary current circuit, leading to a rapid rise in current terminal temperature and fires in secondary current circuit cables and terminal boxes. The system uses infrared thermal imaging to collect real-time infrared thermal imaging data, visible light image data, and temperature and humidity data. An IoT wireless router uploads the collected data to an intelligent analysis platform for analysis and calculation. This identifies the operating status of the secondary circuit of the current transformer within the monitored area and provides timely and appropriate operational alerts. This solves the problems of heavy manual monitoring workload, missed temperature measurement points, and significant influence of ambient temperature on temperature measurement results in current monitoring methods for secondary circuits of current transformers, making timely and accurate fault detection difficult. Monitoring by measuring terminal block temperature only reflects the loosening or detachment of secondary circuit terminals and cannot reflect the type and severity of insulation defects such as aging or damage to the insulation layer, or grounding defects, making it difficult to comprehensively and reliably assess the insulation condition of the secondary circuit. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method and system for monitoring the insulation of secondary AC circuits based on multimodal machine learning. When insulation defects occur in the secondary circuit, partial discharge will occur, and the partial discharge will generate a high-frequency pulse current signal. The insulation status of the secondary circuit can be analyzed by extracting the fault signal features.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for monitoring the insulation of secondary AC circuits based on multimodal machine learning, comprising the following steps:

[0008] S1. The high-frequency pulse current signal generated by the insulation defect in the secondary circuit is processed by multivariate variational mode decomposition to obtain the time-frequency domain characteristics of the fault signal under different modes.

[0009] S2. Select the skewness and kurtosis of the time-domain signal of the high-frequency pulse current, as well as the amplitude and frequency distribution of the frequency-domain signal, as the feature parameters for fault diagnosis. Use the local linear embedding algorithm to reduce the dimensionality of the feature parameters of the fault time-domain reconstructed signal and the fault frequency-domain reconstructed signal in step S1.

[0010] S3. Use the dimensionality-reduced historical fault time-frequency feature vectors as the training set and the dimensionality-reduced real-time fault time-frequency feature vectors as the test set. Classify them using the support vector machine algorithm to obtain the fault types and the fault rates of each fault type.

[0011] In a preferred embodiment, in step S1, the time-domain signal of the high-frequency pulse current signal is first subjected to multivariate variational mode decomposition, and then the different mode time-domain signals generated after decomposition are processed by fast Fourier transform to obtain the corresponding frequency-domain modes.

[0012] In a preferred embodiment, step S1 includes the following steps:

[0013] S101. Use the secondary circuit fault time-domain signal X' as the input signal and set the initial conditions. , ,in For modal parameters, As a penalty factor;

[0014] S102. Perform multivariate variational mode decomposition on the fault time-domain signal X' to obtain... K Each intrinsic mode function is superimposed to form the recombined signal Y';

[0015] S103. Calculate the Pearson correlation coefficient between the fault time-domain signal X' and the reconstructed signal Y', and determine the magnitude of the Pearson correlation coefficient relative to 0.95. The expression for the Pearson correlation coefficient is:

[0016] ;

[0017] in, This represents the Pearson correlation coefficient. and Let X' and Y' represent the standard deviations of the fault time-domain signal and the reconstructed signal, respectively. Let X' be the covariance between the fault time-domain signal X' and the reconstructed signal Y'.

[0018] S104, when At that time, the fault time-domain reconstructed signal Y1 is output, and the corresponding fault frequency-domain reconstructed signal Y2 is obtained by using fast Fourier transform for each eigenmode function of the reconstructed signal Y1.

[0019] like The fault time-domain signal X' is reprocessed using multivariate variational mode decomposition until... .

[0020] In a preferred embodiment, step S2 includes the following steps:

[0021] S201. Select the time-frequency characteristic parameters of the insulation defect in the secondary AC circuit: the skewness of the time-domain signal. Sk The expression is:

[0022] ;

[0023] kurtosis of time-domain signals Ku The expression is:

[0024] ;

[0025] in, For time-domain signal samples, The mean of the time-domain signal samples. n This represents the number of time-domain signal samples.

[0026] S202. Dimensionality reduction of fault feature parameters using a local linear embedding algorithm: N The initial eigenvector matrix is ​​composed of the initial time-frequency characteristic parameters of each mode. The fault time-frequency feature vector composed of D feature parameters is obtained by dimensionality reduction through the local linear embedding algorithm. ,in, express N The initial time-frequency characteristic parameters of each mode, This indicates that the dimensionality reduction process yields... D One time-frequency characteristic parameter;

[0027] The calculation process is as follows:

[0028] 1) Calculate neighboring points:

[0029] Calculate the Euclidean distance between any two sample points using the spatial distance metric formula, and select the closest calculated distance to the sample point. K The Euclidean distance between any two sample points is expressed as follows: (where n points are considered as neighbors).

[0030] ;

[0031] in, Let Euclidean distance be the distance between two sample points. , For any selected sample point, D for D A set of sample points in Euclidean space. p= 2 is a constant;

[0032] 2) Mapping sample points to a low-dimensional space:

[0033] The reconstruction error is calculated using the formula for minimizing reconstruction error. The expression for the formula for minimizing reconstruction error is as follows:

[0034] ;

[0035] in, It is a sample of K Neighboring points, for and The weights between them Z i For the first i The local covariance matrix of each sample point w i For the first i Local reconstruction weight vectors for each sample;

[0036] Z i The expression is:

[0037] ;

[0038] w i The expression is:

[0039] ;

[0040] 3) Calculate the reconstructed weight matrix:

[0041] When mapping all sample points to a low-dimensional space, the following conditions must be met:

[0042] ;

[0043] in, For the output function, For the sample The output vector, for of k Neighboring points, Meet the conditions , , for N×N A sparse matrix used to store , for The i List, for N×N The identity matrix of the first i List, , This represents the output feature vector. These are the final time-frequency feature parameters of the fault extracted after dimensionality reduction.

[0044] In the preferred embodiment, step three, which involves classification using a support vector machine algorithm, includes the following steps:

[0045] S301. Feature Mapping: A kernel function is introduced to map linearly inseparable fault feature data to a high-dimensional space, achieving linear classification. The kernel function satisfies the following conditions:

[0046] ;

[0047] in x , z It is a point in the input space. k For kernel function, For mapping functions;

[0048] S302. Calculate the hyperplane: Find a hyperplane in a high-dimensional space such that the distance from the time-frequency characteristic data points of various faults to the hyperplane is maximized.

[0049] S303, Feature Classification: Obtain the classification results of various fault feature data, and obtain the real-time fault rate by comparing with historical fault feature data.

[0050] This invention also provides a secondary AC circuit insulation monitoring system based on multimodal machine learning, including a high-frequency current transformer, a data acquisition unit, a data aggregation terminal, and a secondary circuit insulation analysis information system. The high-frequency current transformer is installed on phases A, B, and C of the secondary circuit to collect high-frequency pulse current signals generated when insulation defects occur in the secondary AC circuit of a power plant. After receiving the high-frequency pulse current signal, the data acquisition unit wirelessly transmits the data signal to the data aggregation terminal. The data aggregation terminal transmits the collected data from multiple unit cabinets to the secondary circuit insulation analysis information system. The secondary circuit insulation analysis information system executes the aforementioned secondary AC circuit insulation monitoring method based on multimodal machine learning for data processing and fault diagnosis.

[0051] The present invention provides a method and system for monitoring the insulation of secondary AC circuits based on multimodal machine learning, which has the following beneficial effects:

[0052] 1. Quickly locate insulation faults in secondary circuits: Utilizing the characteristic that high-frequency pulse signals are generated when insulation defects occur in secondary circuits, high-frequency current transformers are installed on each secondary circuit to be monitored. When a fault occurs, the corresponding faulty line can be quickly located, making it easier for inspection personnel to troubleshoot and resolve the fault.

[0053] 2. In-depth mining of fault time-frequency domain information: Using the MVMD algorithm and FFT, the high-frequency current signal of insulation defect fault is decomposed into multiple modes in the time and frequency domains. Through the different characteristics of each mode, the characteristics of the fault signal can be analyzed in depth and from multiple perspectives, making fault identification more realistic and reliable.

[0054] 3. Comprehensive and reliable monitoring of secondary circuit insulation status: After performing LLE dimensionality reduction processing on various time-frequency domain characteristic parameters and SVM feature classification, insulation defect fault types such as open circuit, grounding short circuit, insulation layer aging and damage, and loose or detached terminals can be obtained. When a certain insulation defect occurs, the fault type can be accurately identified and the failure rate of that fault type can be obtained through monitoring. Attached Figure Description

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0056] Figure 1 This is a flowchart of the method of the present invention;

[0057] Figure 2 This is a diagram of the secondary AC circuit insulation monitoring system of the present invention;

[0058] Figure 3 This is a flowchart illustrating the technical process of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0060] Example 1:

[0061] This invention provides a method for monitoring the insulation of secondary AC circuits based on multimodal machine learning, such as... Figure 1 and Figure 3 As shown, it includes:

[0062] S1. First, the time-frequency domain of the high-frequency current signal in the secondary circuit is processed by the multivariate variational mode decomposition (MVMD) algorithm to obtain the time-frequency domain characteristics of the fault signal under different modes.

[0063] Time-frequency MVMD decomposition is performed on the high-frequency pulse current signal generated by insulation defects in the secondary circuit. Directly performing Fast Fourier Transform (FFT) on the signal can only reflect the entire frequency domain characteristics of the signal over all time periods. However, by first performing MVMD decomposition on the time domain signal of the high-frequency current, and then performing FFT processing on the different mode time domain signals generated after decomposition to obtain the corresponding frequency domain modes, multi-scale analysis of the fault signal can be performed.

[0064] The basic principle of FFT is to use the Discrete Fourier Transform formula to convert a time-domain signal into a frequency-domain signal. The formula is as follows:

[0065] ;

[0066] in, It is a frequency domain signal. It is the time-domain signal corresponding to the frequency-domain signal. Represents the amplitude of a frequency domain signal. These represent different frequency components.

[0067] The specific process is as follows:

[0068] 1) Using the power plant secondary circuit fault time-domain signal X' as the input signal, set the initial conditions. , ,in For modal parameters, As a penalty factor, when MVMD has the strongest adaptability and the best noise reduction effect.

[0069] 2) Perform MVMD on the fault time-domain signal X' to obtain... K Each intrinsic mode function (IMF) is superimposed to form a recombined signal Y';

[0070] 3) Calculate the Pearson correlation coefficient between the fault time-domain signal X' and the reconstructed signal Y' to determine... With a size of 0.95, where The Pearson correlation coefficient is used. Standard deviation, For covariance;

[0071] 4) When When, output fault time-domain reconstruction signal Y1, if The fault time-domain signal X' is reprocessed using MVMD until... ;

[0072] Simultaneously, each IMF of the reconstructed signal Y1 is used to obtain the corresponding fault frequency domain reconstructed signal Y2 through FFT.

[0073] The reconstructed signal Y' is selected by calculating the Pearson correlation coefficient (correlation coefficient is an indicator of the degree of linear correlation between two variables) with the fault time-domain signal X', and then selecting the reconstructed signal Y' with a high degree of correlation with the fault time-domain signal X', when the following conditions are met: When the condition is met, these recombined signals Y' are taken as Y1.

[0074] S2. Select the skewness and kurtosis of the time-domain signal and the amplitude and frequency distribution of the frequency-domain signal as feature parameters for fault diagnosis. Use the Local Linear Embedding (LLE) algorithm to reduce the dimensionality of the feature parameters of the fault time-domain reconstructed signal Y1 and the fault frequency-domain reconstructed signal Y2 in S1.

[0075] S201. Select the time-frequency characteristic parameters of the insulation defect in the secondary AC circuit. When an insulation defect occurs in the secondary AC circuit, partial discharge will occur. The high-frequency pulse current generated by the partial discharge is a non-stationary transient random signal, which is often mathematically represented by skewness. and kurtosis To describe the characteristics of the distribution of random variables, the high-frequency pulse current signal of partial discharge is considered as a discrete signal of time-amplitude. Therefore, skewness and kurtosis are selected as time-domain characteristic parameters for fault diagnosis. Since different fault types and fault severity correspond to different fault spectrum diagrams, each with corresponding spectral characteristics, frequency content and frequency distribution are selected as frequency-domain characteristic parameters for fault diagnosis.

[0076] S202. Dimensionality reduction of fault characteristic parameters using the LLE algorithm. The initial characteristic parameters of the high-frequency pulse signal of partial discharge are not globally linear. Since the LLE algorithm has the advantages of global optimum without iteration, high computational efficiency, and few variable parameters, it is used to reduce the dimensionality of fault characteristic parameters.

[0077] Will N The initial eigenvector matrix is ​​composed of the initial time-frequency characteristic parameters of each mode. The fault time-frequency feature vector composed of D feature parameters is obtained by dimensionality reduction using the LLE algorithm. ,

[0078] in, express N The initial time-frequency characteristic parameters of each mode, This indicates that the dimensionality reduction process yields... D One time-frequency characteristic parameter.

[0079] The calculation process is as follows:

[0080] 1) Calculate neighboring points:

[0081] According to the formula for measuring spatial distance Calculate the Euclidean distance between any two sample points, and select the one that is closest to the sample point obtained from the calculation. K Each point is considered a neighboring point.

[0082] in, Let Euclidean distance be the distance between two sample points. , For any selected sample point, D for D A set of sample points in Euclidean space. p= 2 is a constant.

[0083] 2) Mapping sample points to a low-dimensional space

[0084] According to the formula for minimizing reconstruction error Perform calculations, where According to the sample of K Neighboring points, for and The weights between them, For the first i The local covariance matrix of each sample point For the first i The local reconstruction weight vector of each sample.

[0085] in, It is a sample of K Neighboring points, for and The weights between them Z i For the first i The local covariance matrix of each sample point, w i For the first i The local reconstruction weight vector of each sample.

[0086] 3) Calculate the reconstructed weight matrix

[0087] When mapping all sample points to a low-dimensional space, the following conditions must be met:

[0088] ;

[0089] in, For the output function, For the sample The output vector, for of k Neighboring points, Meet the conditions , , for N×N A sparse matrix used to store , for The i List, for N×N The identity matrix of the first i List, , This represents the output feature vector. These are the final time-frequency feature parameters of the fault extracted after dimensionality reduction.

[0090] S3. Use the massive historical fault time-frequency feature vectors after dimensionality reduction as the training set and the real-time fault time-frequency feature vectors after dimensionality reduction as the test set. Train the system using the support vector machine (SVM) algorithm and obtain the fault types and the fault rates of each fault type after classification.

[0091] The process of fault classification using the S301 and SVM algorithms is as follows:

[0092] 1) Feature mapping: To avoid overfitting and improve generalization performance, a kernel function is introduced to map linearly inseparable fault feature data to a high-dimensional space, achieving linear classification. The kernel function satisfies the following conditions. ,in x , z It is a point in the input space. k For kernel function, For mapping functions;

[0093] 2) Calculate the hyperplane: Find a hyperplane in the high-dimensional space such that the distance from the time-frequency characteristic data points of various faults to the hyperplane is maximized;

[0094] 3) Feature classification: The classification results of fault feature data such as open wire grounding short circuit, insulation layer aging and damage, and loose and detached terminal are obtained. The real-time fault rate is obtained by comparing with historical fault feature data.

[0095] Example 2:

[0096] This invention also provides a secondary AC circuit insulation monitoring system based on multimodal machine learning, such as... Figure 2 As shown, the system includes: high-frequency current transformers (HFCTs), acquisition units, aggregation terminals, and a secondary circuit insulation analysis information system. The HFCTs are installed on phases A, B, and C of the critical secondary circuits to collect high-frequency pulse current signals generated when insulation defects occur in the power plant's secondary AC circuits. The frequency of the current signal is 3MHz to 30MHz, while low-frequency signals such as the fundamental frequency are filtered out to avoid interference from low-frequency signals to the high-frequency pulse current signal. Each HFCT monitors a corresponding secondary circuit, thus enabling the location of faulty lines. Data is transmitted to the acquisition unit via RS485 communication. After receiving the high-frequency pulse current signal, the acquisition unit wirelessly transmits the data signal to the aggregation terminal. The aggregation terminal then transmits the collected data from multiple unit cabinets to the secondary circuit insulation analysis information system for data processing and fault diagnosis.

[0097] This invention can solve the problem of the lack of single technology for monitoring secondary circuit insulation defects, the problem of conventional power frequency signal analysis (the pulse current signal of insulation defect is mainly concentrated in the high frequency range of 3~30MHz), the problem of the inability of single-angle signal analysis to deeply mine fault signal feature data from multiple angles, and the problem that conventional technology cannot effectively identify the type of secondary circuit insulation defect (open circuit to ground short circuit, insulation layer aging and damage, loose and detached terminals) and the failure rate.

[0098] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A method for monitoring the insulation of secondary AC circuits based on multimodal machine learning, characterized in that, Comprising the following steps: S1, the high-frequency pulse current signal generated by the insulation defect fault of the secondary circuit is processed by multi-element variational modal decomposition to obtain the time-frequency domain characteristics of the fault signal under different modes, first, the time domain signal of the high-frequency pulse current signal is subjected to multi-element variational modal decomposition, and then the time domain signals of different modes generated after decomposition are subjected to fast Fourier transform processing to obtain the corresponding frequency domain mode; Comprising the following steps: S101. Use the secondary circuit fault time-domain signal X' as the input signal and set the initial conditions. , ,in For modal parameters, As a penalty factor; In S102, the multi-element variational modal decomposition is performed on the fault time domain signal X', and decomposition is performed to obtain K a plurality of intrinsic modal functions, and superposition of each intrinsic modal function is a recombined signal Y'. S103, calculate the Pearson correlation coefficient of the fault time domain signal X' and the recombined signal Y', judge the size of the Pearson correlation coefficient and 0.95, the expression of the Pearson correlation coefficient is: ； wherein, represents a Pearson correlation coefficient, and respectively represent the standard deviation of the fault time-domain signal X' and the reorganized signal Y', is the covariance of the fault time-domain signal X' and the reorganized signal Y'. S104、when the fault time-domain reconstruction signal Y1 is output, and meanwhile each intrinsic mode function of the reconstruction signal Y1 is obtained by using fast Fourier transform to obtain the corresponding fault frequency-domain reconstruction signal Y2. If , re-multivariate variational modal decomposition processing is performed on the fault time domain signal X' until ; S2, select the skewness and kurtosis of the time domain signal of the high-frequency pulse current and the amplitude and frequency distribution of the frequency domain signal as the characteristic parameters of fault diagnosis, and use the local linear embedding algorithm to reduce the dimension of the characteristic parameters of the fault time domain reconstruction signal and the fault frequency domain reconstruction signal in step S1; S3, the historical fault time-frequency feature vector after dimension reduction is taken as the training set, and the real-time fault time-frequency feature vector after dimension reduction is taken as the test set, and the support vector machine algorithm is used for classification, and the fault type and the fault rate of each fault type are obtained after classification, the fault type includes broken wire grounding short circuit, insulation layer aging damage and loose terminal falling off.

2. The method of claim 1, wherein the method is based on a multi-modal machine learning. In the step S2, the following steps are included: S201、Select the time-frequency characteristic parameters of the insulation defect of the secondary alternating current loop: skewness of the time domain signal Sk The expression is: ； kurtosis of a time-domain signal Ku The expression for the kurtosis of a time-domain signal is: ； wherein is a time-domain signal sample, is a mean value of the time-domain signal samples, n is a number of time-domain signal samples; S202. Dimensionality reduction of fault feature parameters using a local linear embedding algorithm: N The initial eigenvector matrix is ​​composed of the initial time-frequency characteristic parameters of each mode. The fault time-frequency feature vector composed of D feature parameters is obtained by dimensionality reduction through the local linear embedding algorithm. ,in, express N The initial time-frequency characteristic parameters of each mode, This indicates that the dimensionality reduction process yields... D One time-frequency characteristic parameter; The calculation process is: 1) Calculate the adjacent points: The Euclidean distance between any two sample points is calculated according to the metric space distance formula, and the nearest point to the sample point obtained by calculation is selected as the adjacent point, and the expression of the Euclidean distance between any two sample points is: K ​ ； wherein, is the Euclidean distance between two sample points, , is an arbitrary sample point selected, D is the Euclidean distance between two sample points, D is a set of sample points in a d-dimensional Euclidean space, p= 2 is a constant; 2) Map the sample points to the low-dimensional space: According to the minimum reconstruction error formula, the expression of the minimum reconstruction error formula is: ； wherein is a sample of K neighboring points, is and a weight between Z i is a local covariance matrix of the i sample point, w i is a local reconstruction weight vector of the i sample. Z i The expression is: ； w i The expression is: ； 3) Calculate the reconstruction weight matrix: When mapping all sample points to the low-dimensional space, the condition needs to be met: ； wherein, is an output function, is a sample is an output vector, is is k a neighboring point, satisfies the condition , , is a sparse matrix of N×N is used to store , is the column of i , is the N×N column of the identity matrix, i , , represents the feature vector of the output, is the final fault time-frequency feature parameter extracted after dimension reduction.

3. The method of claim 1, wherein the method is based on a multi-modal machine learning. In the step S3, the support vector machine algorithm is used for classification, including the following steps: S301, feature mapping: introduce a kernel function to map the linearly inseparable fault feature data to a high-dimensional space to realize linear classification, and the kernel function satisfies the condition: ； wherein, x , z is a point in the input space, k is a kernel function, is a mapping function; S302, calculate the hyperplane: find a hyperplane in the high-dimensional space so that the distance from each fault time-frequency feature data point to the hyperplane is maximum; S303, feature classification: obtain the classification results of various fault feature data, and obtain the fault rate of real-time fault by comparing the historical fault feature data.

4. A multi-modal machine learning based secondary AC circuit insulation monitoring system, characterized by, Comprise high-frequency current transformer, acquisition unit, convergence terminal and secondary circuit insulation analysis information system, high-frequency current transformer is installed on A phase, B phase and C phase of secondary circuit, for collecting high-frequency pulse current signal generated when insulation defect occurs in secondary AC circuit of power plant;After receiving the pulse current high-frequency signal, the acquisition unit transmits the data signal to the convergence terminal in a wireless manner;The convergence terminal transes the data of the acquisition unit of the collected multiple unit screens to the secondary circuit insulation analysis information system;The secondary circuit insulation analysis information system executes the secondary AC circuit insulation monitoring method based on multi-modal machine learning in any one of claims 1-3, and carries out data processing and fault diagnosis.

Images