MB2C: Model-Based Deep Reinforcement Learning for Multi-zone Building Control

Reinforcement learning has been widely studied for controlling Heating, Ventilation, and Air conditioning (HVAC) systems. Most of the existing works are focused on Model-Free Reinforcement Learning (MFRL), which learns an agent by extensively trial-and-error interaction with a real building. However, one of the fundamental problems with MFRL is the very large amount of training data required to converge to acceptable performance. Although simulation models have been used to generate sufficient training data to accelerate the training process, MFRL needs a high-fidelity building model for simulation, which is also hard to calibrate. As a result, Model-Based Reinforcement Learning (MBRL) has been used for HVAC control. While MBRL schemes can achieve excellent sample efficiency (i.e. less training data), they often lag behind model-free approaches in terms of asymptotic control performance (i.e. high energy savings while meeting occupants' thermal comfort).

In this paper, we conduct a set of experiments to analyze the limitations of current MBRL-based HVAC control methods, in terms of model uncertainty and controller effectiveness. Using the lessons learned, we develop MB2C, a novel MBRL-based HVAC control system that can achieve high control performance with excellent sample efficiency. MB2C learns the building dynamics by employing an ensemble of environment-conditioned neural networks. It then applies a new control method, Model Predictive Path Integral (MPPI), for HVAC control. It produces candidate action sequences by using an importance sampling weighted algorithm that scales better to high state and action dimensions of multi-zone buildings. We evaluate MB2C using EnergyPlus simulations in a five-zone office building. The results show that MB2C can achieve 8.23% more energy savings compared to the state-of-the-art MBRL solution while maintaining similar thermal comfort. MB2C can reduce the training data set by an order of magnitude (10.52×) while achieving comparable performance to MFRL approaches.