Estimating the insulation health index of XLPE cables using machine learning

This paper focuses on the development of a robust machine learning-based health index model specifically for medium voltage (MV) underground cables insulated with cross-linked polyethylene (XLPE). Maintaining the health and reliability of MV cables is crucial for effective power distribution, as failures can lead to financial losses and outages. Traditionally, maintenance has been reactive, addressing issues only after failures occur, which is costly and results in unnecessary service interruptions. To address this issue, this study introduces a predictive health index model developed using machine learning (ML) techniques. The model aims to shift maintenance strategies from reactive to proactive by identifying cables at risk of failure before they become critical. The dataset used to train the model include various parameters such as environmental factors (moisture, temperature), cable characteristics (age, insulation integrity), and historical performance data (load, partial discharges). By analyzing this dataset, ML algorithms-including Support Vector Machine (SVM), K-Nearest Neighbors (K-NN), Artificial Neural Networks (ANN), and Naïve Bayes-were used to classify cables into distinct health categories, ranging from healthy to at-risk. The results indicate that ANN outperforms conventional methods, achieving 94.5% accuracy-a 12% improvement. Compared to reactive maintenance, this model improves fault prediction by 30% and reduces maintenance costs by 25%.