Exploratory analysis of massive data for distribution fault diagnosis in smart grids

Fault diagnosis in power distribution systems is critical to expedite the restoration of service and improve the reliability. With power grids becoming smarter, more and more data beyond utility outage database are available for fault cause identification. This paper introduces basic methodologies to integrate and analyze data from different sources. Geographic information system (GIS) provides a framework to integrate these data through spatial and temporal relations. Features extracted from raw data provide different discriminant powers, which can be evaluated by the likelihood measure. A fault cause classifier is then trained to learn the relations between fault causes and the features. Two statistical methods, linear discriminant analysis (LDA) and logistic regression (LR), are introduced. The assumptions, general approaches and performances of these two techniques are discussed and evaluated on a real-world outage dataset.