Dueling Double Q-learning based Real-time Energy Dispatch in Grid-connected Microgrids

This paper presents a real-time scheduling strategy based on deep reinforcement learning (DRL) algorithm aiming to realize economic dispatch of microgrid energy storage considering operational uncertainties. Making the scheduling decision of microgrid is a non-trivial task due to the random fluctuations of new energy power generation systems and loads. In order to solve this problem, the double deep Q-learning algorithm with the dueling structure is investigated to ensure the reliability of the microgrid while considering the real-time electricity prices. The agent is tested on the actual data and the results show that the proposed algorithm can get small operation cost of the microgrid in complex situations.