Defect recognition and condition assessment of epoxy insulators in gas insulated switchgear based on multi-information fusion

• A multi-information fusion recognition method based on SDAE and LDDC is proposed. • The feature information is screened by combining M-RAR and SVM. • The concept of evidence credibility was introduced to improve D-S evidence theory. Partial discharge (PD) is one of the main reasons of insulation deterioration in power equipment. How to efficiently and accurately recognize as well as assess the PD is an important guarantee for the stable operation of power equipment. Aiming at the problems of low accuracy of single information defect recognition and condition assessment results, the PD experiments about different typical defects of epoxy insulators are carried out in this paper. On this basis, different ultra high frequency (UHF) and ultrasonic feature information are extracted from time resolved partial discharge (TRPD) and phase resolved partial discharge (PRPD) analysis modes. Furthermore, an SDAE-LDDC multi-information fusion algorithm is built to recognize the defect types, with excellent recognition accuracy (over 94%). Based on the pulse current signal, the severity of PD was classified. The feature information that successfully assesses the severity of PD is selected by combining the maximum correlation and minimal redundancy criterion (m-RMR) with support vector machine (SVM). The performance of selected feature information is tested by fuzzy comprehensive evaluation algorithm, and the assessment results of different feature information are combined using the improved Dempster/Shafer (D-S) evidence theory, which improves the accuracy of condition assessment.