REMOTE DETECTION AND RECOGNITION OF ELECTROSTATIC DISCHARGE FROM HVDC TRANSMISSION LINES

Abstract

To remotely detect corona discharge from High-Voltage Direct Current (HVDC) transmission lines, a detecting system combining detecting platform and data progressing system is designed. Detecting platform is developed resorting to the principle of differential noise reduction, which can fulfill narrow-band detection breaking away interference from broadcasting and easily catch the electrostatic discharge signal. To get rid of interference from spark discharge, a data progressing system containing feature extractions, clustering and recognition technologies is developed. Clustering is realized by extracting five discharge features, including peak factor, form factor, skewness, kurtosis and mean square error. The unsupervised clustering Fuzzy C-Means (FCM) method is used to achieve fast separation for electrostatic discharges and provide training set for pattern recognition. Pattern recognition resorts to Support Vector Machine (SVM) method. For comparison, Back Propagation (BP) and Learning Vector Quantization (LVQ) approaches are taken to test the recognition ability. The results show that SVM recognizer with a recognition rate of 97.5% achieves higher performance than BP and LVQ methods. It can be concluded that the detecting system can be an interesting alternative for electrostatic discharge detection.