Data-driven Sparsity-promoting Optimal Control of Power Buffers in DC Microgrids

A power buffer is a power electronics converter with a large capacitor that shields a weak DC grid from abrupt load changes. Distributed control solutions have been shown to be superior to the decentralized ones, however, the effects of the communication network topology on the control performance of these buffers have not yet been studied. This paper offers a data-driven optimal solution to reduce the interactions between different control loops of power buffers while minimizing a closed-loop performance function. Reinforcement learning methods deal with the optimal control of non-linear systems, and a TabuSearch method addresses the resulting combinatorial problem. The proposed solutions are validated for a DC microgrid in a controller/hardware-in-the-loop environment.