Anomaly Detection in Smart Meter Data for Preventing Potential Smart Grid Imbalance

Abstract

The households and buildings use almost one-third of the total energy consumption among all the power consumption sources. This trend is continuing to rise as more and more buildings install smart meter sensors and connect to Smart Grids and Micro Grids. Smart Grids use sensors and ICT technologies to prevent outages, power imbalance and minimize power wastage. Faults in appliances (like air conditioner duct leakage), abnormal appliances usage (like leaving heating iron, stoves on after usage), sensor faults and abnormal consumer behavior can lead to power outages. Studying the power consumption pattern of houses can lead to a substantial reduction in power wastage which can save millions of dollars. Research works also show that detecting such anomalies can result in preventing outages and save around 20% of power. In this work, we propose an anomaly detection approach for smart meter data for an open data set of houses from Ausgrid Corporation Australia, which is the largest distributor of electricity on Australia’s east coast, providing power to 1.8 million consumers. The power consumption of a house is affected by various factors such as weather and temperature conditions, daily, weekly, yearly seasonality and, holidays. We propose an efficient machine learning-based algorithm to forecast and label power data with anomalies in the first part of this paper. In the second part, after generating the data set with anomaly labels, an efficient machine learning based classification method is proposed to classify power consumption data as either anomalous or normal. We achieve a G-mean score of 97.3% for the proposed classification algorithm. The run time of these classification models is also measured which is within 70 seconds. We performed our experiments on a low capacity Fog device rather than on a Cloud server.