In this paper, a novel optimal learning algo- rithm for partially unknown voltage-source inverters (VSIs) operating in parallel is presented. The algorithm designs game-theory-based distributed controllers to provide the appropriate working voltage magnitude and frequency at the load by converting dc voltage to ac voltage at the par- allel VSIs. It takes advantage of information from the neigh- boring low pass L–C filters to improve harmonic distortion and guarantee equal sharing of the load current across the VSIs while avoiding current circulation during transient and ensuring stability and robustness. It builds upon the ideas of approximate dynamic programming (ADP) and uses only partial information of the system and the exosystem, which is connected only to some of the VSIs. The proposed frame- work was tested in simulations to show its effectiveness.