Detection of consumer electrical load profile anomalies

Abstract

Faulty meters, billing inaccuracies and fraudulent activity all contribute to the non-technical losses experienced by power utilities. Apart from the intended advantage of billing efficiency, the advent of smart meters for consumption and load profile logging, has made possible the analysis of subscriber data for load balancing and consumer classification, amongst other uses. This study analyses data from smart meters in Malta and develops a method for anomaly detection aiming to identify costumers contributing to non-technical losses. Consumers are grouped into clusters with similar weekly consumption patterns using hierarchical clustering. For each consumer, two novel coefficients are then computed: the Anomaly Coefficient, which measures the divergence in consumption patterns for a consumer from all other member of the same cluster, and the Cluster Change Coefficient, which quantifies the likelihood of a customer to exhibit patterns associated with other clusters. Together these two measures are joined in a single Anomaly Score, thus identifying consumers with highly irregular consumption patterns.