Enhanced blind source separation algorithm for partial discharge signals using Joint Approximate diagonalization of Eigenmatrices

Partial discharge (PD) is an early indicator of insulation defects in power equipment. Multiple PD sources create mixed signals, complicating identification. This study simulates mixed PD signals using four typical PD types from Gas Insulated Switchgear (GIS) and proposes an improved Joint Approximate Diagonalization of Eigenmatrices (JADE) algorithm with an amplitude correction coefficient matrix. This method effectively separates mixed PD signals, addressing sign and amplitude uncertainties in traditional blind source separation algorithms. Performance is evaluated using similarity coefficients and root mean square relative errors, with separated signals achieving similarity coefficients above 0.99 and a maximum root mean square relative error of 6.0898. The method demonstrates strong anti-interference capabilities for noisy signals with a signal-to-noise ratio above 5 dB. For measured PD model signals, the average similarity coefficient and root mean square relative error are 0.9998 and 2.8599, respectively.