Conference Paper

Proceedings of Building Simulation 2021: 17th Conference of IBPSA

     

An Open-AI gym environment for the Building Optimization Testing (BOPTEST) framework

Javier Arroyo 1,2,3, Carlo Manna 2,3, Fred Spiessens 2,3, Lieve Helsen 1,2
1 Department of Mechanical Engineering, KU Leuven, Heverlee, Belgium
2 EnergyVille, Thor Park, Waterschei, Belgium
3 Flemish Institute for Technological Research (VITO), Mol, Belgium

DOI: https://doi.org/10.26868/25222708.2021.30380
Abstract: The conventional controllers for building energy management have shown significant room for improvement, and disagree with the superb developments in state-of-the-art technologies like machine learning. This paper describes an OpenAI-Gym environment for the BOPTEST framework to rigorously benchmark different reinforcement learning algorithms among themselves and against other controllers (e.g. model predictive control) by building simulation. The design philosophy of the environment and its different features are introduced. Finally, the environment is demonstrated in one emulator building model to train a reinforcement learning algorithm and compare it against a classical control logic.
Keywords: reinforcement learning, BOPTEST, OpenAI-Gym, building energy management
Pages: 175 - 182
Paper:

bs2021_30380