Machine Learning for Fault Detection and Localization in Underground Power Cables: Improving Reliability in Power Systems

Underground power cables are essential for most power systems today but can experience insulation damage, short-circuit, and mechanical damage. Correct fault detection and accurate fault location is critical in order to reduce outage time and maintenance expenses. This paper proposes a new method for fault diagnosis of the underground power cables using IoT sensors, cloud computing, and ML techniques. The sensors send the voltage, current sensors and soil chemical variaations alongside temperature readings to the ThinkSpeak where they undergo preliminary processing. Three different Random Forest, DenseNet, and the customized pretrained transfer learning model are trained individually to predict and localize the faults using the preprocessed data. Random Forest globally predicts with ensemble decision trees, and DenseNet accurately identifies relationships between parameters if they are nonlinear. A number of Transfer Learning mechanisms are pertinent in analysing time series data and improving the fault prediction results. Different to hybrid systems, this study benchmarks each model and provides a comparison on their effectiveness. The presented results show the efficiency of the introduced framework for high accurate faults detection Random Forest 97%, DenseNet 93%, and transfer learning 99%. That is why based on IoT, cloud computing, and sophisticated ML approaches, this study offers a solution for real-time fault detection of the underground power cables, which can help to develop an intelligent power system.