A Novel Approach to Identify the Anomaly Detection in Electricity usage based on Machine Learning Algorithms and Big Data

Abstract

Comprehension mistakes and frauds may both be detected via the analysis of advanced metering information. By means of on-site examinations or even other techniques, the energy suppliers will be made aware of any suspect usage patterns by this study. The analysis of usage patterns using Machine Learning (ML) techniques may reveal errors, measuring interruptions from assaults, or fine motor control with utility companies. Another of the basic instances of an anomaly detection is fraud detection since it is difficult to categories transactional or utilization information. Additionally, scams vary in type, making it difficult to consistently learn from them. This research study presents an unsupervised ML method to identify out-of-the-ordinary qualities in time series, set a threshold for the proportion of out-of-the-ordinary readings towards the overall measurements, and then classify the moving average as questionable or not. Initially, two different methods are provided to find the differences in time-series based data streaming: Spectral Residual-Convolutional Neural Network and anomaly trained model based on martingales. The properly established information is then subjected to Fisher Linear Discriminant analysis and Two-Class Boosted Decision Tree applications.