An ensemble learning framework for anomaly detection in building energy consumption

Authors

Daniel B. Araya, Western UniversityFollow
Katarina Grolinger, Western UniversityFollow
Hany F. ElYamany, Western UniversityFollow
Miriam AM Capretz, Western University
Girma T. Bitsuamlak, Western University

Document Type

Article

Publication Date

2017

Volume

144

Journal

Energy and Buildings

First Page

191

URL with Digital Object Identifier

http://doi.org/10.1016/j.enbuild.2017.02.058

Last Page

206

Abstract

During building operation, a significant amount of energy is wasted due to equipment and human-related faults. To reduce waste, today's smart buildings monitor energy usage with the aim of identifying abnormal consumption behaviour and notifying the building manager to implement appropriate energy-saving procedures. To this end, this research proposes a new pattern-based anomaly classifier, the collective contextual anomaly detection using sliding window (CCAD-SW) framework. The CCAD-SW framework identifies anomalous consumption patterns using overlapping sliding windows. To enhance the anomaly detection capacity of the CCAD-SW, this research also proposes the ensemble anomaly detection (EAD) framework. The EAD is a generic framework that combines several anomaly detection classifiers using majority voting. To ensure diversity of anomaly classifiers, the EAD is implemented by combining pattern-based (e.g., CCAD-SW) and prediction-based anomaly classifiers. The research was evaluated using real-world data provided by Powersmiths, located in Brampton, Ontario, Canada. Results show that the EAD framework improved the sensitivity of the CCAD-SW by 3.6% and reduced false alarm rate by 2.7%.