Circuit breaker closing vibration signal state evaluation method and system based on multi-feature extraction and multi-classifier fusion

Abstract

A circuit breaker closing vibration signal state evaluation method and system based on multi-feature extraction and multi-classifier fusion are characterized in that the method comprises the following steps: extracting closing signal features from three dimensions of a time domain, a frequency domain and a time-frequency domain, and comprehensively applying a plurality of feature selection methods to calculate feature importance scores; determining the number of finally selected features according to cumulative distribution of feature importance, and performing fault classification on the finally selected features by adopting a multi-classifier optimization integration framework; a high-dimensional feature space is reduced to a 2D or 3D space through a t-SNE technology, and space division of different fault types is realized.

Description

Technical Field

[0001] The present invention relates to the field of power equipment fault assessment, and particularly to a method and system for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion. Background Art

[0002] The closing operation of a circuit breaker is a crucial process, and its quality directly affects the safety and reliability of the operation of the power system. By analyzing the vibration signal generated during the closing process, potential fault problems of the equipment can be effectively identified and evaluated.

[0003] The existing fault assessment methods for closing signals mainly have the following problems: 1. Most are mainly based on single-time domain or frequency domain analysis, lacking multi-dimensional feature extraction of signals, resulting in insufficient evaluation accuracy; 2. Most use a single classification method for fault discrimination and cannot make full use of the complementary advantages of multiple classifiers; 3. The feature extraction and fault classification processes lack an adaptive optimization mechanism and are difficult to cope with complex and variable working conditions; 4. The analysis of feature importance is insufficient, and key features cannot be identified for different fault types.

[0004] Therefore, there is an urgent need for a method for evaluating the vibration signal during the closing process that can comprehensively extract signal features, integrate the advantages of multiple classifiers, and adaptively optimize feature selection to improve the accuracy and reliability of model evaluation. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the present invention provides a method and system for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion.

[0006] The present invention adopts the following technical solutions.

[0007] In a first aspect of the present invention, there is provided a method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion, the method comprising the following steps:

[0008] Extract closing signal features from three dimensions of time domain, frequency domain, and time-frequency domain, and comprehensively apply a variety of feature selection methods to calculate feature importance scores;

[0009] Determine the final selected number of features according to the cumulative distribution of feature importance, and use a multi-classifier optimization integration framework to classify faults for the finally selected features;

[0010] Reduce the high-dimensional feature space to a 2D or 3D space through t-SNE technology to achieve spatial partitioning of different fault types.

[0011] The extracting closing signal features from three dimensions of time domain, frequency domain, and time-frequency domain, and comprehensively applying a variety of feature selection methods to calculate feature importance scores, includes:

[0012] The characteristics of the closing signal extracted in the time domain include basic statistical characteristics, waveform characteristics, morphological characteristics, and distribution characteristics;

[0013] The characteristics of the closing signal extracted in the frequency domain include spectral statistical characteristics, frequency characteristics, and frequency band energy distribution;

[0014] The characteristics of the closing signal extracted in the time-frequency domain include wavelet energy characteristics and wavelet entropy characteristics.

[0015] The extraction of the closing signal characteristics from three dimensions of time domain, frequency domain, and time-frequency domain, and the comprehensive application of various feature selection methods to calculate the feature importance scores, including:

[0016] Based on ANOVA, evaluate the one-way analysis of variance value between each feature and the target variable, and perform analysis of variance on the j-th feature X j and the class label y to obtain the feature importance score S ANOVA (j);

[0017] Use the decision tree model to calculate the importance of each feature, and obtain the importance score S of feature j Tree (j);

[0018] Based on feature selection using mutual information, calculate the mutual information amount between each feature and the target variable, and obtain the importance score S of feature j MI (j);

[0019] Perform weighted fusion on the scoring results of the three methods to obtain S combined (j).

[0020] Determine the number of finally selected features according to the cumulative distribution of feature importance, and use the multi-classifier optimized integration framework to perform fault classification on the finally selected features, including:

[0021] Sort S combined (j) in descending order, calculate the cumulative importance, and determine the number of finally selected features according to the cumulative importance.

[0022] Determine the number of finally selected features according to the cumulative distribution of feature importance, and use the multi-classifier optimized integration framework to perform fault classification on the finally selected features, including:

[0023] Construct the best classifier according to the optimization of the support vector machine classifier, decision tree classifier, KNN classifier, and random forest classifier.

[0024] The construction of the best classifier according to the optimization of the support vector machine classifier, decision tree classifier, KNN classifier, and random forest classifier includes:

[0025] The optimal classifier is

[0026]

[0027] In the formula, ACC classifier is the accuracy of the support vector machine classifier optimization SVM, decision tree classifier optimization Tree, KNN classifier optimization KNN, and random forest classifier optimization RF.

[0028] The high-dimensional feature space is reduced to a 2D or 3D space through t-SNE technology to achieve spatial partitioning of different fault types, including:

[0029] Construct the objective function:

[0030]

[0031] In the formula, P and Q are the conditional probability distributions in the high-dimensional space and the low-dimensional space, respectively, and

[0032]

[0033] Optimize the objective function, and divide the recognition situation and easily confused categories of various faults through the confusion matrix.

[0034] Calculate and visualize the precision, recall rate, and F1 score of various faults to optimize the objective function.

[0035] The confusion matrix is:

[0036]

[0037] In the formula, the confusion matrix C ij represents the number of samples whose true class is i but are predicted as class j.

[0038] In the second aspect of the present invention, it relates to a circuit breaker closing vibration signal state evaluation system based on multi-feature extraction and multi-classifier fusion. The system is implemented by using the method of the first aspect of the present invention; the system includes a multi-dimensional feature extraction module, an optimized feature selection module, a multi-classifier integration module, and a visualization analysis module; the multi-dimensional feature extraction module and the optimized feature selection module extract closing signal features from three dimensions of time domain, frequency domain, and time-frequency domain, and comprehensively apply a variety of feature selection methods to calculate the feature importance score; the multi-classifier integration module determines the finally selected number of features according to the cumulative distribution of feature importance, and uses a multi-classifier optimization integration framework to perform fault classification on the finally selected features; the visualization analysis module reduces the high-dimensional feature space to a 2D or 3D space through t-SNE technology to achieve spatial partitioning of different fault types.

[0039] In a third aspect of the present invention, there is provided a terminal, comprising a processor and a storage medium; the storage medium is used for storing instructions; the processor is used for operating according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0040] In a fourth aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in the first aspect of the present invention are implemented.

[0041] The beneficial effects of the present invention are as follows. Compared with the prior art, a method and system for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion in the present invention can implement the state monitoring and fault warning of power system equipment by extracting rich signal features and adopting an optimized feature selection strategy and a multi-classifier comprehensive evaluation mechanism.

[0042] The beneficial effects of the present invention further include:

[0043] 1. The present invention aims to solve the deficiencies of the existing closing signal fault evaluation methods in feature extraction and classification recognition, and proposes a method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion. By extracting rich signal features and adopting an optimized feature selection strategy and a multi-classifier comprehensive evaluation mechanism, the method can significantly improve the accuracy, robustness and interpretability of fault evaluation, and provide more reliable technical support for the state monitoring and fault warning of power system equipment.

[0044] 2. Feature extraction is carried out from multiple dimensions, and a feature selection method based on multi-index comprehensive evaluation is proposed. Through three-dimensional feature extraction in the time domain - frequency domain - time-frequency domain, the completeness of signal feature representation is improved, and the evaluation accuracy is improved compared with the single feature extraction method. By integrating the advantages of three methods of univariate feature selection, decision tree-based feature selection and mutual information-based feature selection, the most representative feature subset is selected.

[0045] 3. A multi-classifier optimization integration framework is designed. At the same time, four classifiers, namely SVM, decision tree, KNN and random forest, are used for fault identification, and the parameters of each classifier are optimized through cross-validation. Finally, the classifier with the best performance is selected as the evaluation result. The feature selection method of multi-index fusion reduces about 80% of redundant features, reduces the computational complexity, and improves the generalization ability of the classification model at the same time. The multi-classifier optimization integration framework can adaptively select the most suitable classifier for the current working condition, and the comprehensive accuracy is improved compared with a single classifier.

[0046] 4. Based on the visualization analysis method of t-SNE and confusion matrix, the interpretability of the fault evaluation result is enhanced, and the intuitive expression and distinction of fault features under different working conditions are realized. Description of the Drawings

[0047] Figure 1 It is a schematic diagram of the signal data waveform of the circuit breaker closing operation under different faults of the present invention;

[0048] Figure 2 It is a schematic diagram of the recognition accuracy rate of the classifier of the present invention;

[0049] Figure 3 It is a schematic diagram of the t-SNE dimensionality reduction result of the present invention;

[0050] Figure 4 It is a schematic diagram of the confusion matrix of the present invention;

[0051] Figure 5 It is a schematic diagram of the signal state evaluation system of the present invention. Specific embodiments

[0052] To make the purpose, technical solutions and advantages of the present invention clearer and more accurate, the technical solutions of the present invention will be described in detail below through multiple specific embodiments. The embodiments adopted by the present invention are only used to explain the present invention and do not limit the content of the present invention.

[0053] In the first aspect of the present invention, it relates to a circuit breaker closing vibration signal state evaluation method based on multi-feature extraction and multi-classifier fusion. The method includes the following steps:

[0054] Step 1, extract the closing signal features from three dimensions of time domain, frequency domain and time-frequency domain, and comprehensively apply a variety of feature selection methods to calculate the feature importance score.

[0055] Extract the closing signal features from three dimensions of time domain, frequency domain and time-frequency domain, and comprehensively apply a variety of feature selection methods to calculate the feature importance score, including: The closing signal features extracted from the time domain include basic statistical features, waveform features, morphological features and distribution features; The closing signal features extracted from the frequency domain include spectral statistical features, frequency features and frequency band energy distribution; The closing signal features extracted from the time-frequency domain include wavelet energy features and wavelet entropy features.

[0056] The present invention extracts the closing signal features from three dimensions of time domain, frequency domain and time-frequency domain, specifically including:

[0057] ① Time domain features

[0058] Basic statistical features: mean, standard deviation (std), maximum (max), minimum (min)

[0059] Waveform features: peak, peak-to-peak (peakToPeak), root mean square (rms)

[0060] Morphological features: waveformFactor, impulseFactor, marginFactor

[0061] Distribution features: skewness, kurtosis

[0062] ② Frequency-domain feature extraction

[0063] Regarding the frequency-domain characteristics of the closing signal, first convert the time-domain signal to the frequency domain through the Fast Fourier Transform (FFT), and then extract the following features:

[0064] Spectrum statistical features: fftMean, fftStd, fftMax

[0065] Frequency feature: fftPeakFreq

[0066] Frequency band energy distribution: Calculate the energy proportion of each of the four frequency bands (band1Energy - band4Energy) respectively

[0067] ③ Time-frequency feature extraction

[0068] Regarding the non-stationary characteristics of the closing signal, perform time-frequency analysis using wavelet transform and extract the following features:

[0069] Wavelet energy features: approximate coefficient energy (waveletEnergyApprox) and detailed coefficient energy of each layer (waveletEnergyLevel1 - waveletEnergyLevel4)

[0070] Wavelet entropy feature: waveletEntropy

[0071] MATLAB code exists for these feature quantities.

[0072] Extract the closing signal features from three dimensions: time domain, frequency domain, and time-frequency domain. Comprehensively apply various feature selection methods to calculate the feature importance scores, including: evaluating the one-way analysis of variance value between each feature and the target variable based on ANOVA, performing analysis of variance on the j-th feature X j and the class label y to obtain the feature importance score S ANOVA (j); use the decision tree model to calculate the importance of each feature to obtain the importance score S Tree (j) of feature j; perform feature selection based on mutual information, calculate the mutual information amount between each feature and the target variable to obtain the importance score S MI (j) of feature j; perform weighted fusion on the scoring results of the three methods to obtain Scombined (j).

[0073] Univariate feature selection (based on ANOVA): Evaluate the p-value of the one-way ANOVA between each feature and the target variable. j Perform variance analysis with the category label y and get the p value p j , feature importance score S ANOVA (j) is:

[0074] S ANOVA (j) = 1-p j

[0075] Among them, p j is the p calculated in the analysis of variance.

[0076] Feature selection based on decision tree; use decision tree model to calculate the importance of each feature, the importance score of feature j S Tree (j) is defined as:

[0077]

[0078] Among them, I Tree (j) is the importance of the jth feature in the decision tree model, is the maximum value of all feature importances.

[0079] Feature selection based on mutual information calculates the mutual information between each feature and the target variable. The mutual information formula is:

[0080]

[0081] Where p(x,y) is X j and y, where p(x) and p(y) are their respective marginal probability distributions.

[0082] Then, the importance score S of feature j is MI (j) is calculated as:

[0083]

[0084] in, It is the maximum mutual information between all features and labels.

[0085] Perform weighted fusion on the scoring results of the three methods:

[0086] S combined (j) = w1·S ANOVA (j)+w2·S Tree (j)+w3·S MI (j)

[0087] Among them, the weight coefficients are \(w1 = 0.3\), \(w2 = 0.4\), and \(w3 = 0.3\), corresponding to the weights of three methods: ANOVA, decision tree, and mutual information respectively.

[0088] Step 2: Determine the number of finally selected features according to the cumulative distribution of feature importance, and use the multi-classifier optimization integration framework to classify faults for the finally selected features.

[0089] Determine the number of finally selected features according to the cumulative distribution of feature importance:

[0090] Sort \(S\) combined in descending order to get \(S\) sorted , and calculate the cumulative importance:

[0091]

[0092] Select the smallest \(k\) value that satisfies \(CI(k)\geq\alpha\) as the number of features, where \(\alpha\) is the cumulative importance threshold, and in this invention, \(\alpha = 0.5\).

[0093] This invention uses a multi-classifier optimization integration framework for fault classification. This framework includes four different types of classifiers and their parameter optimization methods.

[0094] For the support vector machine (SVM) classifier, the cross-validation method is used for optimization to search for the optimal kernel function and penalty parameter. The kernel function \(K\in\{linear, polynomial, rbf\}\), and the penalty parameter \(C\in\{0.1, 1, 10, 100\}\). 5-fold cross-validation is used, and the accuracy is calculated:

[0095]

[0096] Select the optimal parameter combination:

[0097] [[ID=3⑦]]

[0098] Among them, \(ACC\) i (K, C) is the accuracy of the \(i\)-th fold of validation.

[0099] For the decision tree classifier, optimization is carried out by searching for the optimal minimum leaf node size parameter, and the parameter space is defined as:

[0100] minLeafSize \(\in\{1, 5, 10, 20, 50\}\)

[0101] Use cross-validation to calculate the average accuracy of each parameter value, and select the optimal parameter:

[0102]

[0103] The optimization of the K-nearest neighbor (KNN) classifier defines the parameter space by searching for the optimal number of neighbors and distance metric: the number of neighbors \(K\in\{1,3,5,7,11,15\}\), and the distance metric \(minLeafSize\in\{1,5,10\}\). The average accuracy of each set of parameters \((K, D)\) is calculated using cross-validation, and the optimal parameter combination is selected:

[0104]

[0105] The optimization of the random forest classifier defines the parameter space by searching for the optimal number of trees and minimum leaf size: the number of trees \(numTrees\in\{50,100,200,300\}\), and the minimum leaf size \(minLeafSize\in\{1,5,10\}\). The average accuracy of each set of parameters is calculated using cross-validation, and the optimal parameter combination is selected:

[0106]

[0107] The performance of each classifier on the test set is evaluated, and the classifier with the best performance is selected as the final evaluation result:

[0108]

[0109] where \(y\) i is the true label, is the predicted label, \(m\) is the number of test samples, \([\cdot]\) is the indicator function, and the same applies to the other classifiers.

[0110] Best classifier:

[0111]

[0112] In the formula, \(ACC\) classifier is the accuracy of the support vector machine classifier optimization SVM, decision tree classifier optimization Tree, KNN classifier optimization KNN, and random forest classifier optimization RF.

[0113] Step 3, the high-dimensional feature space is reduced to a 2D or 3D space through the t-SNE technique to achieve the spatial division of different fault types.

[0114] The high-dimensional feature space is reduced to a 2D or 3D space through the t-SNE technique to achieve the intuitive distinction of different fault types. The t-SNE algorithm maps the sample points \(X\) in the high-dimensional space to the low-dimensional space \(Y\), and optimizes the maximum relative entropy objective function:

[0115]

[0116] where \(P\) and \(Q\) are the conditional probability distributions in the high-dimensional space and low-dimensional space respectively:

[0117]

[0118]

[0119] Optimize the objective function and divide the recognition status and easily confused categories of various faults through the confusion matrix.

[0120] The confusion matrix is:

[0121]

[0122]

[0123]

[0124] Recall rate of class i:

[0125]

[0126] F1 score of category i:

[0127]

[0128] Where K is the total number of fault categories.

[0129] This example collects signal data of the circuit breaker closing operation under different fault conditions (normal, latch off, spring slightly loose, spring fully loose, half contact finger). Each data file contains signals from 4 channels. The waveform is as follows Figure 1 .

[0130] The feature extraction described above is performed on the same channel (channel 3) under different faults, and its features are shown in the following table:

[0131]

[0132]

[0133] Combining these three dimensions, we extracted a total of 26 features, forming a feature set that comprehensively characterizes the characteristics of the closing signal. Each signal record generates a 26-dimensional feature vector, along with its corresponding operating condition label, forming feature-label data pairs that facilitate subsequent classification analysis.

[0134] Since the 26 features originally extracted may be redundant and correlated, in order to improve classification efficiency and enhance the generalization ability of the model, this example uses a feature selection method based on multi-index fusion to optimize the feature set.

[0135] First, standardize the feature matrix, standardize each eigenvalue to a distribution with a mean of 0 and a standard deviation of 1, and eliminate the influence of different feature dimensions. Then, use three feature selection methods with different principles to evaluate the feature importance respectively:

[0136] (1) Univariate feature selection based on analysis of variance (ANOVA): Perform a one-way analysis of variance for each feature and the working condition label, calculate the p-value and convert it into an importance score (1 - p). The higher the importance score, the stronger the ability of the feature to distinguish different working conditions.

[0137] (2) Feature selection based on decision trees: Train a decision tree model, extract the importance scores of each feature in the model, and perform normalization. The importance of decision tree features reflects the contribution of features in the process of constructing the decision tree.

[0138] (3) Feature selection based on mutual information: Calculate the mutual information between each feature and the working condition label, and perform normalization. The mutual information reflects the statistical correlation between the feature and the label.

[0139] Then, weight and fuse the feature importance scores of the three methods, with weights of 0.3 for the ANOVA method, 0.4 for the decision tree method, and 0.3 for the mutual information method. According to the fused importance scores, calculate the cumulative importance distribution of the features, and select the feature subset with a cumulative importance reaching 50%. After the feature selection process, 9 most representative features are selected from the original 26 features, including:

[0140] Time domain features: peak, kurtosis, and impulseFactor

[0141] Frequency domain features: fftPeakFreq and band1Energy

[0142] Time-frequency domain features: waveletEnergyLevel1, waveletEnergyLevel3, and waveletEntropy

[0143] These 9 features constitute a feature subset with a lower dimension but rich information, which not only retains the key information but also reduces the computational complexity of subsequent classification.

[0144] Based on feature selection, this example uses a multi-classifier optimization and integration framework for fault classification. First, divide the dataset into a training set and a test set at a ratio of 6:4 to ensure that the distribution of various working condition samples in the training set and the test set is basically the same.

[0145] For the training set data, four different types of classifiers were trained in parallel, and the key parameters of each classifier were optimized by the cross-validation method:

[0146] (1) Support Vector Machine (SVM) classifier: The 5-fold cross-validation method was used to evaluate the combinations of different kernel functions (linear kernel, polynomial kernel, and radial basis kernel) and the penalty parameter C (0.1, 1, 10, 100). After parameter optimization, the best parameter combination was determined to be linear and the penalty parameter C = 10.

[0147] (2) Decision tree classifier: The minimum leaf node size parameter (MinLeafSize) was optimized, and five different parameter values of 1, 5, 10, 20, and 50 were tested. Through cross-validation comparison, the best minimum leaf node size was determined to be 1.

[0148] (3) K-Nearest Neighbor (KNN) classifier: The number of neighbors K (1, 3, 5, 7, 11, 15) and the distance metric (Euclidean distance, Manhattan distance, cosine distance) were comprehensively optimized. After cross-validation, the best parameter combination was determined to be K = 1 and cosine distance.

[0149] (4) Random forest classifier: Two key parameters, the number of trees (50, 100, 200, 300) and the minimum leaf node size (1, 5, 10), were optimized. Through cross-validation comparison, the best parameter combination was determined to be 50 trees and a minimum leaf node size of 5.

[0150] The performance of the four classifiers was evaluated on the test set, and the recognition accuracy of each classifier was as Figure 2 .

[0151] The t-SNE algorithm was used to reduce the dimension to 2D and 3D spaces to visually display the distribution of samples under different working conditions. Two different perplexity parameters (perplexity = 30 and 50) were used for comparison to obtain the best visualization effect. The t-SNE dimensionality reduction results are shown as Figure 3 .

[0152] The confusion matrix is as Figure 4 .

[0153] For each fault category, the precision, recall, and F1-score were calculated as follows:

[0154] Half finger: Precision = 100.00%, Recall = 100.00%, F1-score = 1.00

[0155] Spring fully extended: Precision = 92.31%, Recall = 100.00%, F1-score = 0.96

[0156] Spring slightly loose: Precision = 95.45%, Recall = 95.45%, F1 score = 0.95

[0157] Pin Falling Off: Precision = 100.00%, Recall = 95.65%, F1 Score = 0.98

[0158] Normal: Precision = 95.65%, Recall = 95.65%, F1 score = 0.96

[0159] This example fully verifies the effectiveness and advancement of the circuit breaker closing vibration signal status assessment method based on multi-feature extraction and multi-classifier fusion. By extracting features from three dimensions: time domain, frequency domain, and time-frequency domain, combined with a feature selection method based on multi-index fusion and a multi-classifier optimization integration framework, this method achieves a high accuracy of 100% in closing signal fault assessment. At the same time, visualization analysis methods such as t-SNE dimensionality reduction and confusion matrix enhance the interpretability of the evaluation results, making it easier for engineers to understand and apply them. Experimental results show that the method of the present invention can efficiently and accurately identify various types of faults, providing reliable technical support for power equipment status monitoring and fault warning, and has significant engineering application value.

[0160] like Figure 5 The second aspect of the present invention relates to a circuit breaker closing vibration signal status assessment system based on multi-feature extraction and multi-classifier fusion; the system is implemented by the method described in the first aspect of the present invention; the system includes a multi-dimensional feature extraction module, an optimized feature selection module, a multi-classifier integration module and a visualization analysis module; the multi-dimensional feature extraction module and the optimized feature selection module extract closing signal features from three dimensions: time domain, frequency domain and time-frequency domain, and comprehensively apply multiple feature selection methods to calculate feature importance scores; the multi-classifier integration module determines the final number of selected features based on the cumulative distribution of feature importance, and uses a multi-classifier optimization integration framework to classify the final selected features into faults; the visualization analysis module reduces the high-dimensional feature space to 2D or 3D space through the t-SNE technology to achieve spatial division of different fault types.

[0161] A third aspect of the present invention relates to a terminal, comprising a processor and a storage medium; the storage medium is used to store instructions; and the processor is used to operate according to the instructions to execute the steps of the method described in the first aspect of the present invention.

[0162] A fourth aspect of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect of the present invention.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that there are still modifications or equivalent replacements that can be made to the specific implementation manners of the present invention in the technical solutions of the present invention. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for evaluating the state of the closing vibration signal of a circuit breaker based on multi-feature extraction and multi-classifier fusion, characterized in that The method includes the following steps: Extract the closing signal features from three dimensions: time domain, frequency domain, and time-frequency domain, and comprehensively apply a variety of feature selection methods to calculate the feature importance scores; Determine the finally selected number of features according to the cumulative distribution of feature importance, and use a multi-classifier optimized integration framework to classify the faults for the finally selected features; Reduce the high-dimensional feature space to a 2D or 3D space through t-SNE technology to achieve the spatial division of different fault types.

2. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 1, characterized in that: The extracting the closing signal features from three dimensions: time domain, frequency domain, and time-frequency domain, and comprehensively applying a variety of feature selection methods to calculate the feature importance scores, includes: The closing signal features extracted in the time domain include basic statistical features, waveform features, morphological features, and distribution features; The closing signal features extracted in the frequency domain include spectral statistical features, frequency features, and frequency band energy distribution; The closing signal features extracted in the time-frequency domain include wavelet energy features and wavelet entropy features.

3. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 2, characterized in that: The extracting the closing signal features from three dimensions: time domain, frequency domain, and time-frequency domain, and comprehensively applying a variety of feature selection methods to calculate the feature importance scores, includes: Based on ANOVA, evaluate the one-way ANOVA value between each feature and the target variable, and perform an ANOVA on the j-th feature X j and the class label y to obtain the feature importance score S ANOVA (j); Calculate the importance of each feature using a decision tree model to obtain the importance score S of feature j Tree (j); Feature selection based on mutual information calculates the mutual information between each feature and the target variable to obtain the importance score S MI (j); The scoring results of the three methods are weighted and fused to obtain S combined (j).

4. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 3, characterized in that: The determining the finally selected number of features according to the cumulative distribution of feature importance, and using a multi-classifier optimized integration framework to classify the faults for the finally selected features, includes: Sort S combined (j) in descending order, calculate the cumulative importance, and determine the number of finally selected features based on the cumulative importance.

5. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 4, characterized in that: The determining the finally selected number of features according to the cumulative distribution of feature importance, and using a multi-classifier optimized integration framework to classify the faults for the finally selected features, includes: Construct the optimal classifier according to the optimization of the support vector machine classifier, decision tree classifier, KNN classifier, and random forest classifier.

6. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 5, characterized in that: The constructing the optimal classifier according to the optimization of the support vector machine classifier, decision tree classifier, KNN classifier, and random forest classifier, includes: The optimal classifier is Where ACC classifier is the accuracy of the support vector machine classifier optimization SVM, decision tree classifier optimization Tree, KNN classifier optimization KNN, and random forest classifier optimization RF.

7. A method for evaluating the state of a circuit breaker closing vibration signal based on multi-feature extraction and multi-classifier fusion according to claim 6, characterized in that: The reducing the high-dimensional feature space to a 2D or 3D space through t-SNE technology to achieve the spatial division of different fault types, includes: Construct the objective function: In the formula, P and Q are the conditional probability distributions in the high-dimensional space and the low-dimensional space respectively, and Optimize the objective function, and divide the recognition situations and easily confused categories of various faults through the confusion matrix.

8. A method for evaluating the state of the closing vibration signal of a circuit breaker based on multi - feature extraction and multi - classifier fusion, characterized in that: Calculate and visualize the precision, recall, and F1 - score of various faults to optimize the objective function.

9. A method for evaluating the state of the closing vibration signal of a circuit breaker based on multi - feature extraction and multi - classifier fusion according to claim 8, characterized in that: The confusion matrix is: In the formula, the confusion matrix C ij represents the number of samples whose true class is i but are predicted as class j.

10. A system for evaluating the state of the closing vibration signal of a circuit breaker based on multi - feature extraction and multi - classifier fusion, characterized in that: The system is implemented using the method described in any one of claims 1 - 9; The system includes a multi - dimensional feature extraction module, an optimized feature selection module, a multi - classifier integration module, and a visualization analysis module; The multi - dimensional feature extraction module and the optimized feature selection module extract the characteristics of the closing signal from three dimensions: time domain, frequency domain, and time - frequency domain, and comprehensively apply a variety of feature selection methods to calculate the feature importance score; The multi - classifier integration module determines the finally selected number of features according to the cumulative distribution of feature importance, and uses a multi - classifier optimized integration framework to classify the faults of the finally selected features; The visualization analysis module reduces the high - dimensional feature space to a 2D or 3D space through t - SNE technology to achieve the spatial division of different fault types.

11. A terminal, including a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method described in any one of claims 1 - 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method described in any one of claims 1 - 9.

Images