Abstract:
In order to accurately and efficiently diagnose the discharge fault types of electrical equipment and realize the effective dimensionality reduction of multi-dimensional features, a feature selection and dimensionality reduction method based on correlation and between-class difference is proposed in this paper. Firstly, an audible signal acquisition platform for analog discharge is built. The cross wavelet transform is used to analyze the relevant principal components of the signal, to obtain the signal characteristic frequency band and its corresponding discrete wavelet reconstruction time-domain components, and to extract the multi-dimensional features in time domain of different types of discharge acoustic signals. Then, the Pearson correlation coefficient matrix is used to analyze the correlation between features. Combined with the between-class difference and within-class difference, the feature is selected to test the recognition effect. The recognition accuracy of parameter optimized support vector machine (SVM) is taken as the dimension selection criterion, and the final dimension and feature are determined according to the change law of accuracy. Finally, the method in this paper is compared with the traditional dimensionality reduction algorithm, and the influence of different interference modes on the method is explored. The results show that, compared with the traditional dimensionality reduction algorithm, the proposed method in this paper retains the original feature attributes, and the recognition accuracy of the final selected features exceeds 96%, which provides an effective criterion for feature dimensionality reduction.