Hybrid control of interconnected power converters using both expert-driven droop and data-driven reinforcement learning approaches

The purpose of this investigation is to introduce a new open-source software tool which provides the possibility to train data-driven controllers for power converters in decentralized microgrids such that they can be integrated into a realistic environment with presently existing classical controllers such as droop approaches. We present the methods used for developing some of the control systems implemented in a new software simulation tool, ElectricGrid.jl, written in the Julia programming language. The package solves the ordinary differential equations of electrical networks by modelling the networks as a linear timeinvariant system. Two control schemes are shown to be operating together in a hybrid mode, a classical control scheme operating in the DQ0 reference frame and a reinforcement learning agent.