一种基于UMAP的电缆附件典型缺陷局部放电脉冲特征提取技术OA

Aiming at the problems that the technique for separating the partial discharge blind sources of cable accessories depends on high signal to noise ratio(SNR)while current pulse waveform characteristics are poor in applicability,this paper proposes a pulse feature extraction technique based on uniform manifold approximation and projection(UMAP)to realize the separation of partial discharge pulse signal under various noise interference background.The time-frequency spectrum of partial discharge pulse is taken as the object,the global features of are extracted by UMAP algorithm,eliminating the process of defining and selecting features in traditional feature extraction methods.The dimensionality reduction features can effectively characterize and separate different discharge pulses.According to the partial discharge experimental data of typical defects of cable accessories,the proposed method can achieve feature extraction and separation under white noise with-5 dB SNR,random pulse and narrowband interference,effectively eliminating part of the denoising process in the traditional blind source separation process.Compared with the similar manifold approximation algorithms,this UMAP algorithm has low hyper-parameter sensitivity and stable features after dimensionality reduction,which is conducive to the execution of clustering algorithm.