Multi-View Stacking Ensemble for Power Consumption Anomaly Detection in the Context of Industrial Internet of Things

Abstract

Anomaly detection of power consumption, mainly including electricity stealing and unexpected power energy loss, has been one of the essential routine works in power system management and maintenance. With the help of Industrial Internet of Things technologies, power consumption data was aggregated from distributed various power devices. Hence, the power consumption anomaly was able to be detected by machine learning algorithms. In this paper, a three-stage multi-view stacking ensemble (TMSE) machine learning model based on hierarchical time series feature extraction (HTSF) methods are proposed to solve the anomaly detection problem: HTSF is a novel systematic time series feature engineering method to represent the given data numerically and as input data for machine learning algorithms, while TMSE is designed to ensemble meta-models to archive more accurate performance by using multi-view stacking ensemble method. Performance evaluation in real-world data shows that the proposed method outperforms the existing time series feature extraction means and dramatically decreases the time consumed for ensemble learning process.