Abstract
We consider frequency synchronization and voltage restoration of the isolated micro-grids (MG) by distributed reinforcement learning droop control methods. By exploring the data-driven Q-learning algorithm with the adjacent information sharing mechanism, a fully distributed model-free Q-learning-based droop control is adopted for autonomous frequency synchronization and voltage restoration. Since the proposed distributed control is indeed model-free, it is very suitable for plug-and-play operations of isolated MGs if sufficient operation data of MGs are well-collected. To validate the performance of the proposed method, the proposed distributed Q-learning algorithm was implemented on Matlab/Simulink environment. Simulation results of modified IEEE 34-node distribution system can demonstrate the effectiveness of the proposed distributed Q-learning droop control.