Experimental evaluation of model-free reinforcement learning algorithms for continuous HVAC control

Controlling heating, ventilation and air-conditioning (HVAC) systems is crucial to improving demand-side energy efficiency. At the same time, thermodynamics of buildings uncertainties regarding human activities make effective management challenging. While concept model-free reinforcement learning demonstrates various advantages over existing strategies, literature relies heavily on value-based methods that can hardly handle complex HVAC systems. This paper conducts experiments evaluate four actor-critic algorithms in a simulated data centre. The performance evaluation based their ability maintain thermal stability while increasing efficiency adaptability weather dynamics. Because enormous significance practical use, special attention paid Compared model-based controller implemented into EnergyPlus, all applied reduce consumption by at least 10% simultaneously keeping hourly average temperature desired range. Robustness tests terms different reward functions conditions verify these results. With training, we also see smaller trade-off between reduction. Thus, Soft Actor Critic algorithm achieves stable with ten times less than on-policy methods. In this regard, recommend using future experiments, due both its interesting theoretical properties