Small world stratification for distribution fault diagnosis

Abstract

Automated distribution fault diagnosis generally learns from historical faults and only those relevant to the fault events under study should be investigated. From the spatial perspective, using fault events within a small region is preferred in order to focus on the local fault characteristics. However, a small region may not provide sufficient events for an algorithm to make proper inference about the root cause. To cope with this problem, we propose Small World Stratification (SWS) sampling strategy. SWS involves sampling relevant fault events by Geographic Aggregation (GA) and Feature Space Clustering (FSC), and identifying the group of events that should be investigated together. In this paper, we use simulated fault events to demonstrate that SWS is necessary to improve the fault diagnosis performance when we focus on a small local region and FSC is superior to GA when fault characteristics in neighboring regions are different.