Forecasting Partial Discharges of Cable Joints using Weather data

Abstract

Partial discharge (PD) is a symptom of a weak spot in an underground power cable. Additionally, environmental influences are an important factor in cable degradation. We show that PD in underground cable joints can be successfully forecasted using linear machine learning models leveraging historical PDs and weather data. This has potential applications in estimating the remaining life of cable joints, as we can extend the prediction horizon for predictive maintenance models, such as survival analysis models. Additionally, the model error can be monitored for anomaly detection. This study was conducted in collaboration with Alliander, an electricity and gas distribution system operator in the Netherlands.