Finding unusual energy consumption profiles from large scale data

Abstract

Our energy consumption is increasing every year. The new sensor-based smart measurement devices will provide an opportunity to measure this consumption accurately. This study focuses on finding anomalies from time-series consumption data gathered from various buildings operating in the field of grocery and retail business. The K-means clustering algorithm was used to profile different customers based on their consumption patterns. Investigating these profiles with their respective clusters and applying other statistical methods, possible anomalous locations were extracted for further examination for specialists with domain knowledge.