Detection of anomalous consumers based on smart meter data

Abstract

The continuous smart grid development makes the advanced metering infrastructure an essential part of electricity management systems. Smart meters not only provide consumers with more economical and sustainable electricity consumption but also enable the energy supplier to identify suspicious behaviour or meter failure. In this work, a shape-based algorithm that indicates households with abnormal electricity consumption pattern within a given consumer group was proposed. The algorithm was developed under the assumption that the reason for unusual electricity consumption may not only be a meter failure or fraud, but also consumer’s individual preferences and lifestyle. In the presented methodology, five unsupervised anomaly detection methods were used: K Nearest Neighbors, Local Outlier Factor, Principal Component Analysis, Isolation Forest and Histogram Based Outlier Score. Two time series similarity measures were applied: basic Euclidean distance and Dynamic Time Warping, which allows finding the best alignment between two time series. The algorithm’s performance was tested with multiple parameter configurations on five different consumer groups. Additionally, an analysis of the individual types of anomalies and their detectability by the algorithm was performed.