Predicting of Partial Discharge in Medium-Voltage Cables Using Recurrent Neural Network with Long Short-Term Memory Cells after Wavelet Denoising

Conventional offline partial discharge measurement performed on in-service medium-voltage cables are susceptible to non-stationary noise influence. These disruptive interferences are often unpredictable and lead to complications for the operator when evaluating both on and offsite – potentially incurring higher operation cost. Discrete wavelet transform has proven to be effective in managing temporal variability. Neural networks have accordingly been proven to be useful in identifying the non-linear characteristic found in partial discharge. Multi-step analysis is performed through the two aforementioned methods, with the proposed model shown to substantially reduce recognition difficulty and improve current state-of-the-art evaluation processes both on and offsite. The methodology results in overall increased efficiency and cost reduction.