Smart-Grid Monitoring: Enhanced Machine Learning for Cable Diagnostics

Abstract

Recent works have shown the viability of reusing power line communication modems present in the distribution network for cable diagnostics. By integrating machine learning (ML) techniques, power line modems (PLMs) are shown to be capable of automatically detecting, locating, and assessing different types of cable degradations and faults by monitoring and analyzing their estimated channel frequency responses. However, a single ML algorithm is not ideal for all different diagnostics tasks. To aid us in choosing the most suitable ML algorithm(s) for each of the tasks and to make our solution layman accessible, we propose the use of automated ML, which automatically constructs the best ML model from various algorithms and preprocessing techniques for any given diagnostics task. Our proposed diagnostics approach accelerates the practical deployment of PLM-based grid monitoring by providing a ready-to-use solution to utilities that can be applied without detailed domain knowledge of ML operations.