Anomalous energy consumption detection using a Naïve Bayes approach

Abstract

Background: Industrial energy management has emerged as an important component in monitoring energy consumption particularly with the recent trend of migrating towards IR 4.0. The capability to detect anomalies is essential as it serves as a precautionary step for real-time response to mitigate the maximum demand penalty. The purpose of this research was to develop a high accuracy anomalies detection algorithm to identify anomalies in the energy consumption data recorded by a smart meter. Methods: The proposed algorithm utilized supervised and unsupervised machine learning techniques, namely Isolation Forest and Gaussian Naïve Bayes. The data were first labeled by using Isolation Forest to categorize them into normal and abnormal groups. This was followed by Gaussian Naïve Bayes to classify and predict the anomalies of the smart meter reading. Results: These machine learning techniques showed significant accuracy in predicting the anomalies in smart meter readings. The data used were simulated data collected in less than a month with 30-minute reading intervals. The data were divided into testing and validation sets according to a ratio of 7:3. The balanced accuracy score in predicting anomalies for each different smart meter was above 89%. The average precision, average recall and average F1 score for the normal data were 98%, 99% and 98%, respectively. Whereas the corresponding scores for the abnormal data set were 95%, 90% and 92%. Conclusions: The proposed algorithm is a hybrid approach based on Isolation Forest and Gaussian Naïve Bayes and it provided satisfactory accuracy in anomaly electricity consumption detection based on smart meter readings. The study presents a quick and simple method for categorizing energy consumption data as normal or abnormal, which assists in automatically labelling vast datasets of energy consumption readings. The proposed approach establishes a fundamental framework for predicting the occurrence of anomalies in the industrial energy management system.