Paper Conference

Proceedings of eSim 2020: 11th Conference of IBPSA-Canada


Energy demand prediction in smart buildings using advanced machine learning techniques

Desiree Arias-Requejo 1, Carlos J. Alonso-Gonzalez 2, Belarmino Pulido 2, Marcus M. Keane 1
1 Informatics Research Unit for Sustainable Engineering (IRUSE), National University of Ireland, Galway H91 TK33, Ireland
2 Grupo de Sistemas Inteligentes, Departamento de Informática, Universidad de Valladolid, 47011 Valladolid, Spain

Abstract: Reduction of energy consumption is essential to reduce energy waste. This is even more important in the building sector that accounts for 27% of total CO2 emissions in Europe. In this work, we propose to use available data from a smart building in the NUIG campus (Ireland) to generate black-box models for energy demand prediction using advanced machine-learning techniques. In this paper, we present the first step for an accurate estimation of the energy consumption in buildings. Firstly, hierarchical clustering is used to find the most probable system health state. Secondly, the energy demand models for that state are used to estimate the intended energy consumption. Experimental results showed that the energy demand models specific for a state performs better than a general model for all the system states, confirming our initial hypothesis. This work can be the foundation to perform predictive maintenance based on the energy prediction. In the absence of system faults, a deviation in the energy consumption can be related to tear and wear problems, thus prompting the need for maintenance. Consequently, the reduction in energy consumption due to early detection of a degradation problem will also help to reduce maintenance costs.