Intelligent model prediction of fluctuant increase of maximum electric field in XLPE insulation using long short‐term memory network algorithm

Abstract

The electric field distortion due to space charge accumulations plays a significant role in the ageing, degradation and breakdown in failure of HVDC power cables. Currently, limited experimental results of the electric field dominated by space charges are insufficient to diagnose the power cables. This paper proposes an improved long short-term memory network (LSTM) model for predicting the fluctuating maximum electric field (Emax) in cross-linked polyethylene (XLPE) cable insulation. The various Emax data derived from the complex space charge behaviours were measured using the pulsed electroacoustic method. The model uses regularisation and dropout feedback in the LSTM unit, reducing the phenomenon of over-fitting due to the limited data. It enhances the prediction accuracy and ability of long time prediction by improving the prediction of Emax with the non-linear fluctuation. The predicted Emax approaches 190 kV/mm under 150 kV/mm and 60°C after 2 h. The predicted large variation in Emax under 120 kV/mm and 20°C after 4 h ranges from 130 to 160 kV/mm. It indicates high electric stress in the cable insulation during continuous operation. The proposed LSTM model is of great importance to guide the diagnosis of cable degradation in HVDC power cables.