Paper Conference

Proceedings of Building Simulation 2021: 17th Conference of IBPSA


Operational control of earth-to-air heat exchanger using reinforcement learning

Kento Tomoda 1, Yasuyuki Shiraishi 1, Dirk Saelens 2,3
1 The University of Kitakyushu, Fukuoka, Japan
2 KU Leuven, Department of Civil Engineering, Building Physics & Sustainable Design Section, Leuven, Belgium
3 EnergyVille, Genk, Belgium

Abstract: The purpose of this study is to establish optimal control rules for an earth-to-air heat exchanger (EAHE) using reinforcement learning (RL). We validate the RL control method that achieves two objectives: maintaining the heat load of a fresh air handling unit (FAHU) and suppressing the occurrence of condensation in the EAHE. Compared to scheduled control, RL control increases the annual heat load of the FAHU by approximately 1%, but the time-integrated condensation area ratio is suppressed by approximately 75%. Thus, we confirm that it is possible to achieve the two aforementioned objectives simultaneously using RL control.
Keywords: CFD, Reinforcement Learning Control, Earth-to-air heat exchanger system
Pages: 239 - 246