Study on local discharge type identification based on acoustic signal processing and PCA-BP method

Acoustic detection technology is a technology capable of detecting and pinpointing the origins of abnormal acoustic signals stemming from localized discharges in power equipment. Nevertheless, the current iteration of this technology primarily locates the sources but struggles to elucidate their fundamental origins. As such, a comprehensive composition analysis of these detected acoustic signals is imperative. To tackle this challenge, this paper introduces a methodology for identifying distinct types of localized discharge, alongside the construction of an experimental apparatus simulating various local discharge scenarios. The apparatus employs an acoustic sensor array to capture these sounds. Considering the inherent noise within the experimental setting, we apply a noise reduction strategy to purify the acquired acoustic signals before proceeding with their extraction and identification. We then convert the signals from the time domain to the frequency domain using FFT and wavelet transforms, subsequently extracting their characteristic features via the Principal Component Analysis (PCA) algorithm. Finally, the Back Propagation (BP) neural network ingests these condensed feature vectors, enabling the accurate classification of various local discharge types.