Paper Conference

Proceedings of BSO Conference 2020: Fifth Conference of IBPSA-England


Advanced machine learning techniques for predictive maintenance of HVAC subsystems based on energy consumption prediction

Desiree Arias-Requejo, Carlos J. Alonso-Gonzalez, Belarmino Pulido, Marcus M. Keane

Abstract: Building sector is one of the greatest energy consumers, accounting for 40% of worldwide primary energy. In the buildings, Heating, Ventilation, and Air Conditioning (HVAC) systems are major energy consumers, and their inefficient operation generates up to 50% of energy waste. Hence, techniques capable to find anomalies in these systems due to tear and wear could lead to energy and cost savings. Nowadays the presence of smart buildings provides a valuable source of data from their meters that can be used to build data-driven models. In this work we propose to use these data in two different ways. First, we propose to use weather data to identify different working or operation modes in the buildings via hierarchical clustering. Each cluster can have different and more specific models. Second, we propose to build two kind of models for predicting the energy demand. The first one, a baseline model using historical data. The second one, based on recent data. Our guess is that estimations from both models for future energy demand will be different, being the basis for a predictive maintenance policy. We have tested our approach in a smart building in the National University of Ireland at Galway, the Alice Perry Engineering Building, with promising results.
Pages: 269 - 276