Energy Efficient Operation Optimization of Building Air-conditioners via Simulator-assisted Asynchronous Reinforcement Learning
Date:
The presented study explored a reinforcement learning (RL)-based strategy for optimizing the energy-efficient operation of variable refrigerant flow (VRF) air-conditioners in office settings. The research addressed the significant energy consumption of air-conditioning systems, which account for a substantial proportion of building energy usage, and proposed an innovative solution using asynchronous reinforcement learning coupled with detailed building energy simulation models.
The researchers adopted the Asynchronous Advantage Actor-Critic (A3C) algorithm, a policy-based RL approach, to train control policies using EnergyPlus simulation as the environmental simulator. The RL agent optimized energy savings while maintaining thermal comfort by learning from simulated interactions with the environment. A validation simulator, incorporating different weather conditions, was used to evaluate the robustness of the trained control policies.
The findings demonstrated that the RL-based control strategy achieved up to 16.1% energy savings during the cooling season compared to conventional Bang-Bang control strategies, while also improving thermal comfort. Although the performance was slightly lower in the validation simulator due to environmental variability, the RL control strategy showed a satisfactory level of robustness.
This study highlighted the potential of reinforcement learning as a viable tool for energy-efficient building management. Future work will focus on testing the RL strategy in real-world environments and refining reward functions to further enhance system robustness and learning efficiency.