Fault Detection in Smart Grid Networks by Optimizing the Sensor Network for Distributed Decision Guided by Machine Learning

Abstract

A smart grid network allows the existence of distributed power generation units. These units generate power through renewable or non-renewable means and supply it through the distribution networks. A major problem with these distributed power generation units is that they introduce harmonic components and affect power flow, creating high impedance faults (HIF) in the distribution network. HIF detection is difficult because the associated current has a low amplitude, rendering overcurrent safety devices ineffective. Wireless communication is one of the solutions for fault detection and feeder reconfiguration. This proposed work has an effective sensor network employed to determine and localize the HIF faults in the distribution network supporting distribution generation units. Fast Independent Component features are clustered in each area, and a SVM classifier is constructed to recognize faults. The learnt knowledge represented in SVM is converted to decision rules and disseminated into the sensor network nodes for effective distributed detection and localization of faults. Due to distributed detection, faults can be localized in less time. This improves the accuracy of fault detection as well as improves the network performance.