Advanced Smart Grid Monitoring: Intelligent Cable Diagnostics using Neural Networks

Monitoring and control of network constituents are integral aspects of the smart grid. In this paper, we present a technique for monitoring one such network asset, the underground power cables, which are prone to degradation and damages, resulting in possible power outages. We propose an intelligent cable diagnostics solution using neural networks to determine the health of power cables to predict and prevent eventual faults. To this end, we reuse the communication channel state information inherently estimated by power line modems that are envisioned to enable smart grid communications. We advance the state-of-the-art machine learning based cable health monitoring techniques to present an automated diagnostics procedure using neural networks, which eliminates the need to manually extract features during operation. We demonstrate the architecture of our designed feed-forward neural network, the procedures involved in training, validating, and testing data, and the algorithms we use to train our machines. We evaluate our solution for medium voltage distribution network settings and show through simulation results that our method provides accurate diagnosis in detecting, locating, and assessing cable degradations.