Localisation of Partial Discharge in Power Cables Through Multi-Output Convolutional Recurrent Neural Network and Feature Extraction
Joel Yeo, Huifei Jin, Armando Rodrigo Mor, Chau Yuen, Norasage Pattanadech, Wayes Tushar, Tapan Kumar Saha, Chee Seng Ng
Abstract
This paper proposes an algorithmic approach constructed from a convolutional recurrent neural network (CRNN) iterated with examination of extracted features for partial discharge (PD) localisation; tests were conducted offline on medium voltage (MV) power cables. To evaluate the performance of the algorithm, a case study was performed on 7 cables deliberately selected to comprehensively illustrate the difficulties encountered in field testing. The experimental test results prove that the proposed concept is able to identify and localise discharges besmirched with significant quantities of noise. Main contribution of the methodology is the successful automated interpretation of measurements acquired under noisy challenging field constraints.