A Multi-Agent Deep Deterministic Policy Gradient Method for Multi-Zone HVAC Control

As heating, ventilation, and air conditioning (HVAC) systems consume a significant amount of building energy, an intelligent strategy for building HVAC control has great potential to significantly reduce energy costs. Deep Deterministic Policy Gradient (DDPG) has gained traction in home energy management and HVAC control in recent years. However, the traditional DDPG control is usually ineffective in practice because of the complexity of thermal dynamics with inter-effect between rooms in a building. This paper proposes a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm for HVAC control in an occupant-centric manner that takes into account the occupants’ thermal comfort in multi-zones. The performance of the proposed MADDPG algorithm is evaluated and compared with a traditional DDPG algorithm through EnergyPlus. Simulation results demonstrate that the proposed MADDPG algorithm is more effective than the traditional DDPG approach in energy cost reduction while maintaining the occupants’ multi-zonal thermal comfort.