Filter Based Time-Series Anomaly Detection in AMI using AI Approaches

Detecting anomalies in advanced metering infrastructure can lead to identifying illegal cryptocurrency mining and electricity theft. Successful usage of statistical approaches incorporation with AI models motivated us to propose a combined model on the subject of power consumption in the smart grid. In this paper, we used upper/lower filters to detect sudden and continuous changes in customers' power usage. To improve the dynamic nature of the filters and their accuracy, we clustered users based on extracted statistical features, and ran the Genetic Algorithm to find the optimal hyperparameters and tune the filter of each cluster. Finally, we performed our proposed approach on a real dataset of 999 industrial users measured in the last year (1399 H.S. equivalent to 2020-2021 A.D.).