In parallel inverter microgrid, there is a likelihood of inverter-based distributed generators (IBDGs) being connected with different line impedance. This could result in a significant reactive power sharing error (RPSE). This paper presents the fusion of data-driven control with a conventional virtual synchronous generator (VSG) in a bid to minimize the sharing error. Firstly, all state variables associated with the microgrid are sensed and used as input data for a deep reinforcement learning (DRL) agent. Next, the DRL agent, motivated by a unique reward function, is tasked with controlling the IBDGs such that the voltage of each DG remains within desired limits whilst minimizing the sharing error. The trained agent is deployed on a two-inverter microgrid and the performance is evaluated and compared with the traditional control methods.
