Scheduling Strategy for Energy Storage System in Microgrids Employing Deep Reinforcement Learning

Energy storage system (ESS) plays an essential role in microgrids (MGs). By strategically scheduling the charging/discharging states of ESS, the operational cost of MG can be reduced. In this paper, we consider ESS charging and discharging as decision-making behavior to achieve the goal of minimizing operation cost of MG. The ESS scheduling problem is transformed into a Markov decision process (MDP) by defining ESS action, system state, and reward function. We adopt a deep reinforcement learning (DRL) method of Deep Q-Network with Maximum Mean Discrepancy (MMD-DQN) to efficiently search for the optimal ESS scheduling policy. The convergence reward value of our proposed MMD-DQN method is higher than that of the baseline DQN and D3QN methods. The comparative experiment results display that our MMD-DQN approach can significantly reduce the system operation cost by optimizing the ESS scheduling strategy and outperform the strategies obtained by using existing methods.