Decentralized Data-driven Optimal Control for the Microgrid

Recent advances in climate change policies and sustainable energy systems are spurring the widespread deployment of microgrids. The main operational challenge of these systems is the lack of inertia (in islanded mode) that is typically present in the main grid. To preserve the stable operation of the system, efficient control algorithms are necessary. In this paper, we present a novel optimal control algorithm that leverages constructs from machine learning to decouple interactions between various actuating power components in the microgrid. This allows every actuating entity to make control decisions based only on local measurements. As no communication is necessary with our proposed control algorithm, vulnerabilities and delays that are typically associated with communication-based control algorithms are eliminated. Also, in contrast to traditional decentralized techniques (e.g. droop-based) that are myopic in nature, our algorithm allows for the tracking of the original setpoint without any offsets as the global interactions are accounted for by the machine learning component of the proposed algorithm. We demonstrate the performance, stability and scalability of our proposal via practical simulations conducted on a 15-bus microgrid system.