Data analytics-based anomaly detection in smart distribution network

Abstract

Monitoring energy consumption and diagnosing abnormal behavior will enable utilities to introduce strategies to improve system resiliency, stability, and to meet energy efficiency targets. The deployment of advanced metering infrastructure (AMI) enables utilities to collect various raw data from its customers and networks. This paper presents contextual anomaly detection algorithm to detect irregular power consumption and visualize anomaly scores using unsupervised learning algorithm and temporal context generated from meter readings. The proposed algorithm computes an anomaly score for each user by considering historical consumption data. The anomaly score for a user is then adjusted by analyzing other contextual variables such as seasonal variation day of the week and other users with the same historical pattern. The implementation on real-world data set provided by power utility company shows a high performance of the proposed algorithm.