Hierarchical time series feature extraction for power consumption anomaly detection

Abstract

Anomaly of power consumption, particularly due to electricity stealing, has been one of the major concern in power system management for a long time, which may destroy the demand-supply balance and lead to power grid regulating issues and huge profit reduction of electricity companies. One of the essential key to develop machine learning model to solve the above problems is time series feature extraction, which may affect the superior limit of machine learning model. In this paper, a novel systematic time series feature extraction method named hierarchical time series feature extraction is proposed, used for supervised binary classification model that only using user registration information and daily power consumption data, to detect anomaly consumption user with an output of stealing probability. Performance on data of over 100,000 customers shows that the proposed methods are outperforming one of the existing state-of-the-art time series feature extraction library tsfresh [1].