Advanced Sensing Techniques for Cable Fault Detection Enhanced by Power-Line Communications and Integrated Feature-Image Analysis With Residual Neural Networks

Exploiting power line communication (PLC), a distinctive noninvasive sensing system has been developed to detect insulation defects in distribution networks, providing early alerts for impending insulation failures and averting system disruptions. Despite this progress, existing detection algorithms fall short of the precision necessary for real-time grid monitoring due to variations introduced by load dynamics and grid configurations. In this study, we delve into the evolution of cable parameters from water tree degradation to ultimate insulation failure, formulating a comprehensive dielectric model that mirrors the aging process of cable insulation. We introduce an innovative approach that hinges on anomalies classification within feature images. This approach harnesses the Markov transition field (MTF) to encode channel frequency response (CFR) data visually and employs a residual neural network (ResNet), enhanced with an attention mechanism, to pinpoint various cable defects. Remarkably, our proposed methodology boasts a perception accuracy exceeding 98% for minute parameter fluctuations within intricate networks.