GCB arc contact electrical endurance monitoring method based on multi-source signal fusion sensing

Abstract

The invention discloses a GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing, and relates to the technical field of power system monitoring, and the method comprises the steps: obtaining multi-source signal data in the action process of a circuit breaker; performing preprocessing and feature extraction on the multi-source signal data to generate feature parameters of each signal; constructing a multi-information fusion feature set based on the feature parameters; constructing a multi-layer perceptron model, and training the multi-layer perceptron model by using the multi-information fusion feature set; and inputting the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, outputting a classification result of the ablation degree of the GCB arc contact, and completing the monitoring of the electrical life of the GCB arc contact. According to the invention, through comprehensive utilization of the multi-source signal data and construction of the multi-information fusion feature set, high-precision classification of the ablation degree of the circuit breaker arc contact is realized, and the accuracy of electrical life monitoring is effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of power system monitoring, in particular to a GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing. Background Art

[0002] In recent years, the research on online monitoring and fault diagnosis of high-voltage switchgear has mainly focused on single-source signal analysis or independent analysis of multi-source signals, which limits the accuracy of diagnosis and classification of fault types. Multi-parameter information fusion technology can improve the accuracy of fault diagnosis by simulating faults, collecting signals, extracting features, and combining classification algorithms. Especially for high-voltage SF6 circuit breakers, the reliability of their contact systems is crucial to power systems. Traditional electrical life detection methods, such as arc energy-based prediction and dynamic loop resistance methods, have limitations. Therefore, studying non-invasive measurement methods suitable for online monitoring to predict the degree of arc contact ablation and the electrical life of circuit breakers is of great significance for the reliable operation of power systems.

[0003] In previous studies, there are two main types of traditional detection methods: prediction methods based on arc energy and dynamic loop resistance methods. The core idea of ​​the prediction method based on arc energy is that the life of the electrical contact depends on the cumulative arcing energy. This type of method is relatively rough in practical applications. Alternatively, a large number of breaking tests are performed on a certain type of circuit breaker through the N-Ib life curve method to find out the relationship between the breaking life of the circuit breaker under different breaking currents, which can obtain more accurate results. However, this method is expensive and it is difficult to measure the life curve for all types of circuit breakers. The dynamic loop resistance method measures the change in resistance during the opening and closing process of the circuit breaker, which can intuitively reflect the ablation of the arc contact in the arc extinguishing chamber of the circuit breaker. However, it can only be measured when the circuit breaker is out of service, and online monitoring of the electrical life cannot be achieved. Summary of the invention

[0004] In view of the limitations of existing single-source signal analysis and traditional electrical life detection methods, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to utilize multi-source signal fusion technology to optimize the signal processing and feature extraction process to realize the monitoring of the electrical life of the GCB arc contact.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, an embodiment of the present invention provides a GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing, which includes obtaining multi-source signal data during the operation of the circuit breaker, the multi-source signals including the opening and closing coil current signal, vibration signal, travel signal and radio frequency signal; preprocessing and feature extraction of the multi-source signal data to generate feature parameters of each signal; constructing a multi-information fusion feature set based on the feature parameters, including: taking the vibration signal feature parameters as the basis vector, and combining the opening and closing coil current signal feature parameters and the radio frequency signal feature parameters on the basis vector to form a multi-information fusion feature set; constructing a multi-layer perceptron model, and training the multi-layer perceptron model using the multi-information fusion feature set; inputting the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, outputting the classification result of the degree of ablation of the GCB arc contact, and completing the electrical life monitoring of the GCB arc contact.

[0007] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, the preprocessing and feature extraction of multi-source signal data includes the following steps: time synchronization processing of multi-source signal data, establishing a unified time reference system, and applying an adaptive filtering algorithm to perform noise reduction processing on the synchronized multi-source signal data; for the opening and closing coil current signal, the key feature points of the opening and closing coil current signal are identified by wavelet transform and threshold detection method, and mapped to the travel curve; for the vibration signal, S transform is used for time-frequency analysis to extract the vibration signal characteristics, and the vibration signal characteristics include time-frequency energy distribution characteristics and main modal characteristics; for the travel signal, polynomial fitting smoothing processing is performed to extract the travel signal characteristics; for the radio frequency signal, digital envelope technology is used to extract the radio frequency signal characteristics, and the radio frequency signal characteristics include radio frequency envelopes; the nonlinear characteristics of the multi-source signal data are calculated, and the nonlinear characteristics include approximate entropy, sample entropy and multi-scale entropy; the key feature points, vibration signal characteristics, travel signal characteristics, and radio frequency signal characteristics are preliminarily normalized, an initial feature vector is constructed, and characteristic parameters of each signal are generated.

[0008] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, wherein: using S transform for time-frequency analysis includes the following steps: defining a continuous expression of S transform, the specific formula is as follows: , in, is the time shift parameter, f is the signal characteristic frequency, is the vibration signal to be measured, j is the imaginary unit, and t is the time variable; the S transform is discretized to obtain the discrete form of the S transform; the modulus square of the S transform result is calculated to obtain the time-frequency energy spectrum, the time-frequency energy spectrum is normalized, and the time-frequency energy distribution characteristics are extracted. The time-frequency energy distribution characteristics include total energy, frequency band energy ratio, energy center of gravity, and energy standard deviation; for each time point , determine the time-frequency energy spectrum The local maximum in the image is tracked, the change of the local maximum over time is tracked, the time-varying characteristics of the main mode are obtained, and the main mode features are extracted.

[0009] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, wherein: the use of digital envelope technology to extract the radio frequency envelope includes the following steps: recording the input radio frequency signal as , the output envelope signal is recorded as ; Establish envelope extraction rules; discretize the continuous-time envelope extraction rules and express them using difference equations. The specific formula is as follows: , Where n is the discrete time index, fs is the sampling frequency, is the time constant, is the envelope signal, For RF signal; select value, balance the response speed and smoothness of the envelope; initialization , for each time point n, apply the difference equation to calculate , until the entire RF signal is processed.

[0010] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, wherein: constructing a multi-information fusion feature set based on feature parameters includes the following steps: organizing the feature parameters of the vibration signal into basis vectors; adding the feature parameters of the opening and closing coil current signal, the radio frequency signal, and the travel signal to the basis vectors; for each action process of the circuit breaker, executing the feature fusion process, generating multiple fused feature vectors, and combining all the generated fused feature vectors into a feature matrix X; using a recursive feature elimination algorithm to select features of the feature matrix X, while considering the importance of each signal source to the circuit breaker state diagnosis, screening the optimal feature subset; applying the kernel principal component analysis method to the selected optimal feature subset to perform nonlinear feature extraction; normalizing the features after the kernel principal component analysis Processing, organizing the normalized features into the final multi-information fusion feature set; screening the optimal feature subset includes: if the absolute value of the Pearson correlation coefficient between the features is greater than the first preset value, then combining the maximum correlation minimum redundancy algorithm based on mutual information and the RFE algorithm to construct a two-stage feature selection process: in the first stage, the mRMR algorithm is applied to calculate the mutual information value between each feature and the circuit breaker status label, and the features with mutual information values ​​in the top 1 / 2 are selected; the redundancy between the selected features is calculated, and if the calculated feature redundancy is greater than the second preset threshold, the features with the largest average mutual information with other features are eliminated one by one from the selected features until the feature redundancy is less than or equal to the second preset threshold; in the second stage, the RFE algorithm is applied to the selected feature subset to select the features that contribute most to the circuit breaker status diagnosis.

[0011] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, the following steps are included in the construction of a multi-layer perceptron model: designing a multi-layer perceptron model structure, including an input layer, a feature weighting layer, at least one hidden layer, an attention mechanism layer, and an output layer; initializing the multi-layer perceptron model parameters, using an initialization strategy of a unit matrix plus a small random perturbation for the feature weighting layer, and using a He initialization method for the hidden layer; inputting a multi-information fusion feature set into the multi-layer perceptron model, and adaptively weighting features from different sources through a feature weighting layer; using multiple hidden layers to perform nonlinear transformation on the weighted features, wherein the number of hidden layers is dynamically increased according to the complexity of the input features. dynamic adjustment; apply the attention mechanism layer after the last hidden layer to highlight the most relevant features for judging the degree of ablation of GCB arc contacts; generate the classification probability distribution of the degree of ablation of GCB arc contacts through the output layer; calculate the cross entropy loss between the output of the multilayer perceptron model and the true label, and add the L2 regularization term; use the Adam optimizer, combined with the learning rate decay strategy, to update the model parameters based on the calculated loss function value; at the end of each training cycle, evaluate the performance of the multilayer perceptron model and dynamically adjust the weights of the feature weighting layer; repeat the iterative training process until the preset convergence conditions are met; perform sparse processing on the multilayer perceptron model that meets the convergence conditions, obtain the final multilayer perceptron model and save it.

[0012] As a preferred solution of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing described in the present invention, the output layer uses a softmax function to output the classification result of the GCB arc contact ablation degree. The specific formula is as follows: , Where T is the temperature parameter, is the feature representation after being processed by the feature weighting layer, multiple hidden layers and the attention mechanism layer. is the output vector, K is the number of classifications of the ablation degree of the GCB arc contact, is the output probability of the jth class, and e is the natural exponential function.

[0013] In the second aspect, an embodiment of the present invention provides a GCB arc contact electrical life monitoring system based on multi-source signal fusion sensing, which includes a signal acquisition module for acquiring multi-source signal data during the operation of the circuit breaker, the multi-source signals including opening and closing coil current signals, vibration signals, travel signals and radio frequency signals; a preprocessing and feature extraction module for preprocessing and feature extraction of multi-source signal data to generate feature parameters of each signal; a multi-information fusion module for constructing a multi-information fusion feature set based on the feature parameters; a model construction module for constructing a multi-layer perceptron model and training the multi-layer perceptron model using the multi-information fusion feature set; a monitoring classification module for inputting the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, outputting the classification result of the degree of ablation of the GCB arc contact, and completing the electrical life monitoring of the GCB arc contact.

[0014] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as described in the first aspect of the present invention are implemented.

[0015] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as described in the first aspect of the present invention are implemented.

[0016] The beneficial effects of the present invention are as follows: the present invention realizes high-precision classification of the ablation degree of the circuit breaker arc contact through the comprehensive utilization of multi-source signal data and the construction of a multi-information fusion feature set, and effectively improves the accuracy of electrical life monitoring. A variety of signal processing technologies such as adaptive filtering, wavelet transform, S transform and polynomial fitting are used to effectively reduce noise and extract features for different types of signals, which significantly improves the robustness and reliability of the system under complex working conditions. By combining the recursive feature elimination algorithm (RFE) and the maximum relevance minimum redundancy (mRMR) algorithm based on mutual information, the feature matrix is ​​optimized and screened to ensure that the features used in the model training process have the highest discrimination and reduce the computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1The framework flow chart of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing.

[0019] Figure 2 It is the form of various signals of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing.

[0020] Figure 3 It is the vibration signal of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing and its characteristic curve extracted after S transformation.

[0021] Figure 4 This is the shape of the RF signal after envelope in the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing.

[0022] Figure 5 It is a computer equipment diagram of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing. DETAILED DESCRIPTION

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0026] Example 1, reference Figure 1~Figure 5 , which is the first embodiment of the present invention, provides a GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing, including: S1: Acquire multi-source signal data during the circuit breaker operation process.

[0027] It should be noted that the multi-source signal data includes the opening and closing coil current signal, vibration signal, travel signal and radio frequency signal.

[0028] S2: Preprocess and extract features of multi-source signal data to generate feature parameters of each signal; Specifically, the following steps are included: S2.1: Perform time synchronization processing on multi-source signal data, establish a unified time reference system, and apply an adaptive filtering algorithm to perform noise reduction processing on the synchronized multi-source signal data.

[0029] It should be noted that time synchronization ensures that each signal corresponds accurately on the time axis, providing a unified time reference for subsequent analysis; the adaptive filtering algorithm can automatically adjust the filter parameters according to the characteristics of different signals, effectively removing noise from various signals while retaining useful information.

[0030] S2.2: For the closing and opening coil current signal, the key feature points of the closing and opening coil current signal are identified through wavelet transform and threshold detection method, and mapped to the travel curve.

[0031] Specifically, the closing and opening coil current signal is decomposed by wavelet, and the appropriate wavelet basis and decomposition level are selected; the adaptive threshold detection algorithm is applied to the wavelet coefficients, and the signal is reconstructed by inverse transformation to locate and identify the key feature points; the identified key feature points are mapped to the travel curve, such as Figure 2 shown.

[0032] It should be noted that the key characteristic points include the current starting point, inflection point and breaking point.

[0033] S2.3: For vibration signals, S transform is used to perform time-frequency analysis to extract time-frequency energy distribution characteristics and main modal characteristics.

[0034] Preferably, the S transform combines the phase characteristics of the short-time Fourier transform and the multi-resolution analysis capability of the wavelet transform, and can provide good time and frequency resolution at the same time. The continuous expression of the S transform is defined as follows: , in, is the time shift parameter (used to control the position of the Gaussian window on the time axis t), f is the signal characteristic frequency, is the vibration signal to be measured, j is the imaginary unit, and t is the time variable.

[0035] In actual analysis and processing, it is necessary to discretize the S transform, that is, , , the discrete form of S transform is: , Where x is the time index, y is the frequency index, T is the sampling period, N is the number of sampling points, and m is the time index. is the discrete sample of the vibration signal to be measured.

[0036] It should be noted that is a Gaussian window function, which adjusts its width according to the characteristic frequency f of the signal. is the Fourier basis function, is a discrete Gaussian window function, is the discrete Fourier basis function.

[0037] Furthermore, the square of the modulus of the S-transform result is calculated to obtain the time-frequency energy spectrum ; For the time-frequency energy spectrum Normalization processing is performed to extract time-frequency energy distribution characteristics including total energy, frequency band energy ratio, energy center of gravity and energy standard deviation.

[0038] Furthermore, for each time point , find the time-frequency energy spectrum The local maxima in the model correspond to the instantaneous main modes. The changes of these local maxima over time are tracked to obtain the time-varying characteristics of the main modes, and the main mode features are extracted, including the main mode frequency, modal damping ratio and modal participation factor.

[0039] Furthermore, the time-varying curve of the main modal frequency is plotted; the statistical characteristics of the modal frequency (mean, variance, etc.) are calculated; and the time-varying characteristics of the modal participation factor are analyzed to reflect the dynamic changes in the importance of each mode.

[0040] It should be noted that the vibration signal and its characteristic curve extracted after S transformation are as follows: Figure 3 shown.

[0041] S2.4: Perform polynomial fitting and smoothing processing on the travel signal to extract travel features.

[0042] The travel characteristics include the movement start time, maximum speed moment, maximum speed value and average acceleration of the travel signal.

[0043] It should be noted that the order of the polynomial fitting should be selected according to the complexity of the travel signal and the required smoothness to achieve a balance between retaining important features and removing noise.

[0044] S2.5: For RF signals, digital envelope technology is used to extract the RF envelope.

[0045] Specifically, the following steps are included: S2.5.1: Let the input RF signal be , the output envelope signal is recorded as .

[0046] S2.5.2: Establish envelope extraction rules.

[0047] Specifically, when season ;when season ,in represents the most recent time before time t, is the time constant.

[0048] S2.5.3: Discretize the continuous-time envelope extraction rule and express it using a difference equation.

[0049] Specifically, the difference equation is expressed as follows: , Where n is the discrete time index, fs is the sampling frequency, is the time constant, is the envelope signal, For radio frequency signals.

[0050] S2.5.4: Select the appropriate value to balance the response speed and smoothness of the envelope.

[0051] Preferably, The value can be selected from 0.1ms to 10ms, and the sampling frequency fs can be selected from 10kHz to 1MHz.

[0052] It should be noted that the smaller The envelope follows the signal more closely. A higher value produces a smoother envelope, so the RC value can be dynamically adjusted according to the characteristics of the signal to adapt to changes in different frequency components.

[0053] S2.5.5: Initialization , for each time point n, apply the difference equation to calculate , until the entire RF signal is processed.

[0054] Furthermore, the RF signal after the envelope is as follows Figure 4 shown.

[0055] S2.6: Compute nonlinear features of multi-source signal data, including approximate entropy, sample entropy, and multiscale entropy.

[0056] S2.7: Perform preliminary normalization processing on the extracted feature parameters, construct an initial feature vector, and generate feature parameters of each signal.

[0057] It should be noted that the extracted characteristic parameters include the closing and opening coil current signal characteristics (current starting point, inflection point and breaking point), vibration signal characteristics (time-frequency energy distribution characteristics and main modal characteristics), travel signal characteristics (movement start time, maximum speed moment, maximum speed value and average acceleration), RF signal characteristics (key characteristic points of the RF envelope, such as peak value, rise time, fall time), and nonlinear characteristics (approximate entropy, sample entropy and multi-scale entropy).

[0058] S3: Construct a multi-information fusion feature set based on feature parameters.

[0059] Specifically, the following steps are included: S3.1: Organize the characteristic parameters of the vibration signal into basis vectors.

[0060] S3.2: Add the characteristic parameters of the closing and opening coil current signal, radio frequency signal, and travel signal to the basis vector.

[0061] S3.3: For each action process of the circuit breaker, execute the feature fusion process to generate multiple fused feature vectors, and combine all the generated fused feature vectors into a feature matrix X.

[0062] It should be noted that the feature fusion process refers to steps S3.2 and S3.3.

[0063] S3.4: Use the recursive feature elimination algorithm (RFE) to select features from the feature matrix X, and consider the importance of each signal source to the circuit breaker status diagnosis to screen the optimal feature subset.

[0064] Preferably, if there is a strong correlation between the features (i.e., the absolute value of the Pearson correlation coefficient between the features is greater than 0.8), the maximum relevance minimum redundancy (mRMR) algorithm based on mutual information and the RFE algorithm are combined to construct a two-stage feature selection process: in the first stage, the mRMR algorithm is applied to calculate the mutual information value between each feature and the circuit breaker status label, and the features with mutual information values ​​in the top 1 / 2 are selected, and the redundancy between the selected features is calculated. If the calculated feature redundancy is greater than a preset threshold (0.3), the features with the largest average mutual information with other features are eliminated one by one from the selected features until the feature redundancy is less than or equal to the preset threshold (0.3); in the second stage, the RFE algorithm is applied to the selected feature subset to select the features that contribute most to the circuit breaker status diagnosis.

[0065] Specifically, support vector machine is used as the base classifier, and the 10% features with the lowest feature importance scores are removed in each iteration until the number of remaining features reaches the preset target.

[0066] S3.5: Apply the kernel principal component analysis method (KPCA) to the selected optimal feature subset for nonlinear feature extraction.

[0067] Specifically, if the number of samples of the feature matrix X exceeds 10,000 or the number of features exceeds 100, the following method is used: implement the batch KPCA algorithm to divide the data set into batches of size 1,000; perform KPCA on each batch to extract the principal components with an explained variance ratio of 95%; merge the KPCA results of each batch to obtain the final nonlinear feature representation.

[0068] S3.6: Normalize the features after kernel principal component analysis, and organize the normalized features into the final multi-information fusion feature set.

[0069] It should be noted that KPCA can capture the nonlinear relationship between features, while RFE helps to select the most discriminative feature subset. The combination of the two can effectively reduce the dimension of the feature space while retaining key information.

[0070] S4: Build a multi-layer perceptron model and train the multi-layer perceptron model using a multi-information fusion feature set.

[0071] Specifically, the following steps are included: S4.1: Design an improved multi-layer perceptron model structure, including an input layer, a feature weighting layer, multiple hidden layers (including convolutional layers and fully connected layers), an attention mechanism layer, and an output layer.

[0072] Among them, the input layer receives the multi-information fusion feature set, and the output layer uses the softmax function to output the classification result of the ablation degree of the GCB arc contact.

[0073] It should be noted that the specific formula of the Softmax function is as follows: , Where T is the temperature parameter (used to adjust the smoothness of the output distribution), is the feature representation after being processed by the feature weighting layer, multiple hidden layers and the attention mechanism layer. is the output vector, K is the number of classifications of the ablation degree of the GCB arc contact, is the output probability of the jth class, and e is the natural exponential function.

[0074] S4.2: Initialize the parameters of the multilayer perceptron model, use the unit matrix plus small random perturbation initialization strategy for the feature weighting layer, and use the He initialization method for the hidden layer.

[0075] S4.3: Input the multi-information fusion feature set into the multi-layer perceptron model, and adaptively weight the features from different sources through the feature weighting layer.

[0076] S4.4: Use multiple hidden layers to perform nonlinear transformations on the weighted features, where the number of hidden layers is dynamically adjusted based on the complexity of the input features.

[0077] S4.5: An attention mechanism layer is applied after the last hidden layer to highlight the most relevant features for judging the ablation degree of GCB arc contacts.

[0078] S4.6: Generate the classification probability distribution of the GCB arc contact ablation degree through the output layer.

[0079] S4.7: Calculate the cross entropy loss between the multilayer perceptron model output and the true label, and add an L2 regularization term to prevent overfitting.

[0080] It should be noted that the specific formula of the loss function is as follows: , Where L is the total loss, is the cross entropy loss, is the L2 regularization term, is the regularization coefficient, N is the number of samples, K is the number of classifications of the ablation degree of the GCB arc contact, is the true label (0 or 1) of the i-th sample belonging to the j-th class, The model predicts the probability that the i-th sample belongs to the j-th class, W is the set of all weight parameters in the model, and w is a single weight parameter in the model.

[0081] S4.8: Use the Adam optimizer, combined with the learning rate decay strategy, to update the model parameters based on the calculated loss function value.

[0082] S4.9: At the end of each training cycle, evaluate the performance of the multilayer perceptron model and dynamically adjust the weights of the feature weighting layer.

[0083] S4.10: Define a training loop and repeat the iterative training process (S4.2~S4.9) until the preset convergence conditions are met.

[0084] It should be noted that the preset convergence conditions include the following: the loss function value no longer decreases significantly for multiple consecutive epochs (for example, the loss decreases by less than 0.1% for 5 consecutive epochs); the performance indicator (such as accuracy) reaches the preset threshold; and the maximum number of training rounds is reached.

[0085] S4.11: Perform sparse processing on the multilayer perceptron model that meets the convergence conditions to obtain a streamlined multilayer perceptron model and save it.

[0086] S5: Input the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, output the classification result of the ablation degree of the GCB arc contact, and complete the electrical life monitoring of the GCB arc contact.

[0087] Furthermore, the present embodiment also provides a GCB arc contact electrical life monitoring system based on multi-source signal fusion sensing, including a signal acquisition module for acquiring multi-source signal data during the circuit breaker operation process, the multi-source signals including opening and closing coil current signals, vibration signals, travel signals and radio frequency signals; a preprocessing and feature extraction module for preprocessing and feature extraction of multi-source signal data to generate feature parameters of each signal; a multi-information fusion module for constructing a multi-information fusion feature set based on the feature parameters; a model construction module for constructing a multi-layer perceptron model and training the multi-layer perceptron model using the multi-information fusion feature set; a monitoring classification module for inputting the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, outputting the classification result of the degree of ablation of the GCB arc contact, and completing the electrical life monitoring of the GCB arc contact.

[0088] This embodiment also provides a computer device, which is suitable for the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as proposed in the above embodiment.

[0089] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0090] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for monitoring the electrical life of a GCB arc contact based on multi-source signal fusion sensing as proposed in the above embodiment is implemented.

[0091] In summary, the present invention achieves high-precision classification of the degree of ablation of circuit breaker arc contacts through the comprehensive utilization of multi-source signal data and the construction of multi-information fusion feature sets, effectively improving the accuracy of electrical life monitoring. A variety of signal processing technologies such as adaptive filtering, wavelet transform, S transform and polynomial fitting are used to effectively reduce noise and extract features for different types of signals, significantly improving the robustness and reliability of the system under complex working conditions. By combining the recursive feature elimination algorithm (RFE) and the maximum relevance minimum redundancy (mRMR) algorithm based on mutual information, the feature matrix is ​​optimized and screened to ensure that the features used in the model training process have the highest discrimination and reduce the computational complexity.

[0092] Example 2, reference Figure 1~Figure 5 , which is the second embodiment of the present invention, and this embodiment provides a GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0093] Specifically, the experiment selected a 500kV SF6 circuit breaker in a substation as the research object, and collected its operating data from January to December 2023, including multi-source signals such as opening and closing coil current, vibration, travel and radio frequency. The experiment was carried out strictly in accordance with the technical steps from S1 to S5. The following is a detailed experimental process and result analysis.

[0094] First, in the data acquisition stage, a high-precision data acquisition system is used to collect multi-source signals for each action process of the circuit breaker. The sampling rate of the closing and opening coil current signal is 10kHz, with a range of 0-100A; the sampling rate of the vibration signal is 20kHz, with a frequency range of 0-10kHz; the sampling rate of the travel signal is 5kHz, with a range of 0-300mm; the sampling rate of the radio frequency signal is 100MHz, with a frequency range of 300MHz-3GHz. A total of 1,000 circuit breaker action data were collected, of which 800 were used for training and 200 for testing. The duration of each action data was 200ms.

[0095] Furthermore, in the data preprocessing and feature extraction stage, GPS timestamps are used to synchronize all signals with an accuracy of 1μs, and adaptive wavelet threshold denoising algorithm is used for noise reduction, which improves the signal-to-noise ratio by about 5dB. Subsequently, various types of signals are processed in a targeted manner: wavelet transform is used to identify key feature points of the closing and opening coil current signal; S transform is performed on the vibration signal and time-frequency features are extracted; polynomial fitting is performed on the travel signal and motion features are extracted; digital envelope extraction is performed on the RF signal. In addition, nonlinear features of each signal are calculated, such as approximate entropy, sample entropy, and multi-scale entropy.

[0096] Furthermore, a 52-dimensional initial feature vector was constructed. Subsequently, 26 features were selected using the mRMR algorithm, and the RFE algorithm was used to finally select the 18 best features. Next, the KPCA method was used for nonlinear feature extraction, and 5 principal components were extracted, with a cumulative explained variance ratio of 95.3%. Finally, the features were Z-score standardized to obtain an 18-dimensional fused feature vector.

[0097] Furthermore, an improved multi-layer perceptron model was designed, which includes an input layer, a feature weighting layer, multiple hidden layers (including convolutional layers and fully connected layers), an attention mechanism layer, and an output layer. The model uses the ReLU activation function, the Adam optimizer, a batch size of 64, 200 training rounds, and L2 regularization to prevent overfitting. During the training process, the accuracy of the model increased from the initial 72.5% to the final 98.7%, and the validation set accuracy reached 97.2%. After the training was completed, the model was sparsely processed, and the number of parameters was reduced by 32% while the accuracy was only reduced by 0.3%.

[0098] In the model testing and result analysis phase (S5), the trained model was fully evaluated using 200 test data. The test results showed that the overall accuracy of the model reached 96.5%, and the kappa coefficient was 0.956, indicating that the model had excellent classification performance. The precision, recall, and F1 scores of each category were all above 94.9%, further confirming the stability and reliability of the model. The predicted probability distribution of each category was analyzed, and the results are shown in Table 1, which fully demonstrated the model's ability to accurately identify different degrees of ablation.

[0099] Table 1 Output vector table , As shown in Table 1, this experiment successfully verified the effectiveness of the GCB arc contact electrical life monitoring method based on multi-source signal fusion and multi-layer perceptron. This method can accurately identify different degrees of arc contact ablation, providing a reliable basis for the preventive maintenance of circuit breakers. Future research directions include expanding the scale of the data set and optimizing the model structure to improve the generalization ability and practicality of the method, and provide stronger support for the safe and stable operation of the power system.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for monitoring the electrical life of GCB arc contacts based on multi-source signal fusion sensing, characterized in that: include, Acquire multi-source signal data during the circuit breaker operation process, wherein the multi-source signal data includes a closing and opening coil current signal, a vibration signal, a travel signal, and a radio frequency signal; Preprocessing and feature extraction are performed on the multi-source signal data to generate feature parameters of each signal; Constructing a multi-information fusion feature set based on the feature parameters includes: The vibration signal characteristic parameters are used as basis vectors, and the opening and closing coil current signal characteristic parameters and the radio frequency signal characteristic parameters are combined onto the basis vectors to form a multi-information fusion feature set; Constructing a multi-layer perceptron model, and training the multi-layer perceptron model using the multi-information fusion feature set; The preprocessed multi-source signal data to be detected is input into the multi-layer perceptron model, and the classification result of the ablation degree of the GCB arc contact is output to complete the electrical life monitoring of the GCB arc contact.

2. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as claimed in claim 1 is characterized in that: The preprocessing and feature extraction of the multi-source signal data comprises the following steps: Perform time synchronization processing on multi-source signal data, establish a unified time reference system, and apply adaptive filtering algorithm to perform noise reduction processing on the synchronized multi-source signal data; For the closing and opening coil current signal, the key feature points of the closing and opening coil current signal are identified through wavelet transform and threshold detection method, and mapped to the travel curve; For the vibration signal, S transform is used to perform time-frequency analysis to extract the vibration signal features, which include time-frequency energy distribution features and main modal features; For the travel signal, polynomial fitting and smoothing processing is performed to extract the travel signal features; For radio frequency signals, digital envelope technology is used to extract radio frequency signal features, and the radio frequency signal features include radio frequency envelope lines; Calculating nonlinear features of multi-source signal data, wherein the nonlinear features include approximate entropy, sample entropy and multi-scale entropy; The key feature points, the vibration signal features, the travel signal features, and the radio frequency signal features are preliminarily normalized to construct an initial feature vector and generate feature parameters of each signal.

3. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as claimed in claim 2 is characterized in that: The time-frequency analysis using S transform comprises the following steps: Define the continuous expression of S transform, the specific formula is as follows: ， in, is the time shift parameter, f is the signal characteristic frequency, is the vibration signal to be measured, j is the imaginary unit, and t is the time variable; Discretize the S transform to obtain the discrete form of the S transform; Calculate the square of the modulus of the S-transform result to obtain a time-frequency energy spectrum, normalize the time-frequency energy spectrum, and extract time-frequency energy distribution characteristics, wherein the time-frequency energy distribution characteristics include total energy, frequency band energy ratio, energy center of gravity, and energy standard deviation; For each time point , determine the time-frequency energy spectrum The local maximum in the image is tracked, the change of the local maximum over time is tracked, the time-varying characteristics of the main mode are obtained, and the main mode features are extracted.

4. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as claimed in claim 2 is characterized in that: The method of extracting radio frequency signal characteristics by using digital envelope technology comprises the following steps: The input RF signal is recorded as , the output envelope signal is recorded as ; Establish envelope extraction rules; Discretize the continuous-time envelope extraction rule and express it using a difference equation. The specific formula is as follows: ， Where n is the discrete time index, fs is the sampling frequency, is the time constant, is the envelope signal, is a radio frequency signal; choose value, balance the response speed and smoothness of the envelope; initialization , for each time point n, apply the difference equation to calculate , until the entire RF signal is processed.

5. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as claimed in claim 1 is characterized in that: The constructing of a multi-information fusion feature set based on the feature parameters comprises the following steps: Organizing characteristic parameters of vibration signals into basis vectors; Add characteristic parameters of the closing and opening coil current signal, radio frequency signal and travel signal to the basis vector; For each action process of the circuit breaker, the feature fusion process is executed to generate multiple fused feature vectors, and all the generated fused feature vectors are combined into a feature matrix X; The recursive feature elimination algorithm is used to select features from the feature matrix X, and the importance of each signal source to the circuit breaker status diagnosis is considered to screen the optimal feature subset; Apply the kernel principal component analysis method to the selected optimal feature subset for nonlinear feature extraction; Normalize the features after kernel principal component analysis and organize the normalized features into the final multi-information fusion feature set; The screening of the optimal feature subset comprises: If the absolute value of the Pearson correlation coefficient between the features is greater than the first preset value, a two-stage feature selection process is constructed by combining the maximum relevance minimum redundancy algorithm based on mutual information with the RFE algorithm: In the first stage, the mRMR algorithm is applied to calculate the mutual information value between each feature and the circuit breaker status label, and the features with mutual information values ​​in the top 1 / 2 are selected; Calculate the redundancy between the selected features, and if the calculated feature redundancy is greater than a second preset threshold, remove the features with the largest average mutual information values ​​with other features from the selected features one by one until the feature redundancy is less than or equal to the second preset threshold; In the second stage, the RFE algorithm is applied to the selected feature subset to select the features that contribute most to the circuit breaker status diagnosis.

6. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing according to claim 1 is characterized in that: The construction of the multi-layer perceptron model comprises the following steps: Design a multi-layer perceptron model structure, including an input layer, a feature weighting layer, at least one hidden layer, an attention mechanism layer, and an output layer; Initialize the parameters of the multilayer perceptron model, use the unit matrix plus small random perturbation initialization strategy for the feature weighting layer, and use the He initialization method for the hidden layer; The multi-information fusion feature set is input into the multi-layer perceptron model, and the features from different sources are adaptively weighted through the feature weighting layer; Use multiple hidden layers to perform nonlinear transformation on the weighted features, where the number of hidden layers is dynamically adjusted according to the complexity of the input features; An attention mechanism layer is applied after the last hidden layer to highlight the most relevant features for judging the ablation degree of GCB arc contacts; Generate the classification probability distribution of GCB arc contact ablation degree through the output layer; Calculate the cross entropy loss between the multi-layer perceptron model output and the true label, and add an L2 regularization term; Use the Adam optimizer, combined with the learning rate decay strategy, to update the model parameters based on the calculated loss function value; At the end of each training cycle, the performance of the multi-layer perceptron model is evaluated and the weights of the feature weighting layer are dynamically adjusted; Repeat the iterative training process until the preset convergence conditions are met; The multilayer perceptron model that meets the convergence conditions is sparsely processed to obtain the final multilayer perceptron model and save it.

7. The GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as claimed in claim 6 is characterized in that: The output layer outputs the classification result of the ablation degree of the GCB arc contact through the softmax function. The specific formula is as follows: ， Where T is the temperature parameter, is the feature representation after being processed by the feature weighting layer, multiple hidden layers and the attention mechanism layer. is the output vector, K is the number of classifications of the ablation degree of the GCB arc contact, is the output probability of the jth class, and e is the natural exponential function.

8. A GCB arc contact electrical life monitoring system based on multi-source signal fusion sensing, based on the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing according to any one of claims 1 to 7, characterized in that: Also includes, A signal acquisition module is used to obtain multi-source signal data during the circuit breaker operation process, wherein the multi-source signals include a closing coil current signal, a vibration signal, a travel signal, and a radio frequency signal; A preprocessing and feature extraction module, used for preprocessing and feature extraction of the multi-source signal data to generate feature parameters of each signal; A multi-information fusion module, used for constructing a multi-information fusion feature set based on the feature parameters; A model building module, used to build a multi-layer perceptron model, and train the multi-layer perceptron model using the multi-information fusion feature set; The monitoring classification module is used to input the preprocessed multi-source signal data to be detected into the multi-layer perceptron model, output the classification result of the ablation degree of the GCB arc contact, and complete the electrical life monitoring of the GCB arc contact.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the GCB arc contact electrical life monitoring method based on multi-source signal fusion sensing as described in any one of claims 1 to 7 are implemented.

Images