Distributed Q-Learning Secondary Voltage Control of Isolated AC Microgrids with Saturated Input

A model-free data-driven Q-learning algorithm for distributed control in isolated AC MGs is proposed for achieving the autonomous voltage restoration. First, by using the feedback linearization technique to each DG, a second-order linear system is defined. Then the distributed voltage restoration problem is re-formulated a consensus problem. We study this consensus problem subject to saturated inputs. An iterative model-free datadriven Q-learning algorithm is adopted to achieve the voltage restoration. Simulation of an isolated micro-grids have performed to demonstrate the effectiveness of the proposed method.