Supplementary material to the paper. Includes datasets, charts and additional code.
The experiments were conducted using Sinergym, a framework for building energy simulation.
The building and construction sector -and particularly Heating, Ventilation, and Air Conditioning (HVAC) systems- pose a significant challenge to climate change mitigation given its high impact on global energy consumption and CO2 emissions. Recent studies have addressed efficient HVAC control by employing different Deep Reinforcement Learning (DRL) algorithms, but in general they are are explicitly designed for specific setups and do not provide a standardized comparison of their performance. This paper aims to fill this gap by developing a critical and reproducible evaluation of state-of-the-art DRL algorithms applied to HVAC control in various simulated buildings and configurations. To this end, we considered several aspects of energy consumption and occupants' comfort: and took into account the performance of agents, their robustness under different conditions, their capability to adapt from simpler to more complex environments, and the trade-off between optimization goals. We used the Sinergym software, a simulation framework capable of running experiments involving multiple models and configurations. The obtained results confirm the potential of DRL algorithms, such as SAC and TD3, compared to reactive control methods, and reveal interesting insights about their generalization and learning capabilities.
Reinforcement Learning, HVAC, Building Energy Optimization, Sinergym
- Antonio Manjavacas
- Alejandro Campoy-Nieves
- Javier Jiménez-Raboso
- Miguel Molina-Solana
- Juan Gómez-Romero
Department of Computer Science and Artificial Intelligence (DECSAI), Universidad de Granada, Spain.
Sustainable Artificial Intelligence Lab (SAIL).
@article{...}
Click here to view/download the bibliography of the paper.
Read LICENSE.