Mixed Deep Reinforcement Learning Considering Discrete-continuous Hybrid Action Space for Smart Home Energy Management

This paper develops deep reinforcement learning (DRL) algorithms for optimizing the operation of home energy system which consists photovoltaic (PV) panels, battery storage system, and household appliances. Model-free DRL can efficiently handle difficulty modeling uncertainty PV generation. However, discrete-continuous hybrid action space considered challenges existing either discrete actions or continuous actions. Thus, a mixed (MDRL) algorithm is proposed, integrates Q-learning (DQL) deterministic policy gradient (DDPG) algorithm. The DQL deals with actions, while DDPG handles MDRL learns optimal strategy by trial-and-error interactions environment. unsafe violate constraints, give rise to great cost. To such problem, safe-MDRL further proposed. Simulation studies demonstrate that proposed challenge from management. reduces cost maintaining human thermal comfort comparing benchmark on test dataset. Moreover, greatly loss in stage