Fault area location in electrical distribution system using smart meter data

Power distribution systems are becoming increasingly digitalized. This provides a lot of potential for data-driven protection, automation, and maintenance methods. However, large-scale integrated hardware and software solutions can be difficult to manage, so maintenance methods that use data that are already collected are of high value. The contribution of this paper is to propose a new method for fault area location in power distribution systems. The method uses a technique for multivariate process monitoring based on principal component analysis, which employs both fault detection and diagnosis. It uses only aggregated smart meter data, which is normally collected for billing purposes, and improves the features using feature engineering. The output of the method is a list of component labels, which is ordered according to which components are most likely to be affected by an interruption. The method is tested on data from a real Danish power distribution system and can precisely identify, medium-voltage transformers and low-voltage radials that are affected by an interruption and gives good indications of which electrical boxes are affected. The method integrates with a GIS system to provide coordinates of affected components and, importantly, gives no false positives fault detections.