Fast Decision Generation for Building Energy Management System Decisions Based on Behavior Cloning

As the proportion of renewable energy resources increases, their inherent uncertainty demands increased flexibility in power systems. As major power consumers, the heating, ventilation and air conditioning (HVAC) systems in buildings are increasingly being dispatched as demand-side flexible resources. Recent studies mainly focus on model predictive control (MPC) and reinforcement learning (RL). However, the hardware level of most current buildings is insufficient to apply these methods. To tackle this issue, behavior cloning (BC) is used to clone the policies of these advanced approaches with lower hardware demands. Firstly, the MPC is used to obtain the expert demonstration data. Secondly, the learner policy is generated via BC. Finally, a precise HVAC system model is established based on EnergyPlus to validate the effectiveness of the proposed method. Simulation results show the proposed method can achieve efficient control of building HVAC systems with low hardware requirements.