Proactive Learning Renewable Energy Management Strategy for Isolated DC Microgrid

Complex weather conditions are an important factor restricting the development of solar power generation. In this paper, an proactive learning-based dynamic power management method for DC microgrid fuel cell/battery systems is proposed to cope with the uncertainty of solar power generation. A deep deterministic policy gradient algorithm is used to adaptively distribute power between fuel cell/battery systems. Secondly, the demand power curve is randomly generated using the historical data of microgrid PV to ensure the efficiency of the strategy under any photovoltaic power generation conditions. The findings indicate that this energy management strategy notably decreases both battery deterioration and fuel cell costs by 18.6% and 21.06% under PV-abundant scenarios, and by 5.07% and 9.13% under PV-scarce scenarios, respectively. These improvements bring the system’s performance closer to that achieved by offline dynamic programming.