Behavior Anomaly Detection Fused with Features of Mel Frequency Cepstrum Coefficients

Abstract

In recent years, with the rise of artificial intelligence, the research in the field of pattern recognition is developing rapidly. Its supporting algorithms mainly include neural network and support vector machine (SVM), which are widely used in various fields such as industrial detection, aviation, medical treatment, finance, internet and so on. On the one hand, the effect of pattern recognition is related to the selection and optimization of the algorithm, on the other hand, it is closely related to the noise reduction methods to improve the signal quality, the feature extraction and feature selection in feature engineering. Mel frequency cepstrum coefficient (MFCC) is widely used in the field of speech recognition, and has achieved good recognition effect for low-frequency speech signal pattern recognition, but its pattern recognition in ultrasonic field is rarely studied. In this paper, a method of extracting MFCC in ultrasonic frequency range and applying it to SVM pattern recognition and fault prediction is proposed. In the experimental verification stage, GIS-PD model is used to collect data of acoustic emission (AE) signal induced by partial discharge (PD) of gas insulated switchgear (GIS), and MFCC eigenvalues are extracted in the ultrasonic frequency range of 20-200 kHz. Finally, MFCC and optimized selected time-domain and frequency-domain eigenvalues are used for SVM pattern recognition to monitor and warn the running state of GIS equipment. Based on the comprehensive scheme proposed in this study, the overall accuracy rate is improved to ninety percent after fusion of MFCC eigenvalues, which is improved by 22.5 percent compared with only using time-frequency domain eigenvalues. At present, the research on MFCC extraction in the ultrasonic frequency range of 20-200 kHz and its application in SVM pattern recognition has not been reported. This study will promote the research of MFCC in the field of ultrasonic detection, and provide reference for the research in related fields.