A scheme for electricity theft detection based on EWMA control chart

This paper proposes a methodology for electricity theft detection based on Exponentially Weighted Moving Average (EWMA) control chart. EWMA charts monitor customers' consumption data to identify whether they are engaged in malicious activity or not. EWMA charts are quite insensitive to normality assumptions, so they are a powerful tool for analyzing real -world consumption data with any statistical distribution. This methodology uses consumption data available from smart meter installed at consumers' premises only without requiring any central observer meter. Previous research using control charts for theft detection assumed anomaly free historical power consumption data, impacting parameter estimation accuracy in real world datasets. To address this, two robust techniques have been used to estimate mean and standard deviation of consumption data, enhancing resistance of EWMA chart to anomalies. Additionally, the methodology incorporates the nondominated sorting genetic algorithm (NSGA-II) for selecting control chart design parameters, ensuring the optimal combination of decision metrics across all theft levels. The NSGA-II algorithm empowers users to customize control charts, optimizing various decision metrics. The method was tested on two real -world datasets, one with residential customer consumption data and the other with small and medium enterprises (SMEs) consumption data. The results demonstrate strong performance across various metrics.