Abstract
To achieve autonomous frequency synchronization in AC microgrids, an off-policy learning algorithm using an actor-critic structure for distributed generators (DGs) is proposed. The algorithm applies a linear model of P-f droop to each DG and reformulates the DGs in MG as a consensus problem. Reinforcement learning techniques are used to study this consensus problem, and the actor-critic method is adopted to achieve frequency synchronization. The algorithm for the actorcritic method on each DG is presented, and its effectiveness is validated through simulation under an isolated AC microgrid. The proposed algorithm is compared with other methods to demonstrate its superiority. Overall, the actor-critic structure of off-policy learning algorithm shows promising results for achieving frequency synchronization in AC microgrids.