Multi-Agent Deep Reinforcement Learning for HVAC Control in Commercial Buildings

In commercial buildings, about 40%-50% of the total electricity consumption is attributed to Heating, Ventilation, and Air Conditioning (HVAC) systems, which places an economic burden on building operators. this paper, we intend minimize energy cost HVAC system in a multi-zone with consideration random zone occupancy, thermal comfort, indoor air quality comfort. Due existence unknown dynamics models, parameter uncertainties (e.g., outdoor temperature, price, number occupants), spatially temporally coupled constraints associated temperature CO2 concentration, large discrete solution space, non-convex non-separable objective function, it very challenging achieve above aim. To end, minimization problem reformulated as Markov game. Then, control algorithm proposed solve game based multi-agent deep reinforcement learning attention mechanism. The does not require any prior knowledge uncertain parameters can operate without knowing models. Simulation results real-world traces show effectiveness, robustness scalability algorithm.