AdaHeat: A General Adaptive Intelligent Agent for Domestic Heating Control

Abstract

Improving the energy efficiency of domestic heating systems can lead to a major reduction in energy consumption and the corresponding CO2 emissions. To this end, intelligent domestic heating agents (IDHAs) aim to operate domestic heating systems more efficiently with minimum user input. In this work, we propose a new general IDHA that balances heating cost and thermal discomfort in an infinite horizon optimization manner, learns an adaptive thermal model of the system under control on-line and plans a heating schedule that fully exploits the probabilistic occupancy estimates. Importantly, our agent adapts to the user preferences in balancing heating cost and thermal discomfort, as it relies on a single parametrization variable that is learned on-line, and is able to consider a wide range of heating systems typically employed in domestic settings. The backbone of our IDHA is an adaptive model predictive control approach along with a new general planning algorithm that utilizes dynamic programming. We present a thorough evaluation of our approach, and show its effectiveness in terms of Pareto efficiency and usability criteria against state-of-the-art IDHAs. By so doing, we also conduct a comprehensive characterization of existing IDHAs to provide significant insights about their performance in different operational settings.