Cost Effective Dynamic Multi-Microgrid Formulation Method Using Deep Reinforcement Learning

This paper proposes an online Dynamic Multi-Microgrid Formulation (DMMF) method using Deep Reinforcement Learning. It aims to reconfigure the microgrid into several self-supplied islands and minimize total operation cost at the same time. Spanning Tree Algorithm is used to reduce the total number of microgrid formulation. Proximal-Policy optimization is implemented to train the agent which determines the status of sectionalizing switches in microgrid in real-time. To show the effectiveness of the proposed DMMF method, a case study was conducted in the modified cigre-14 bus test network. The results demonstrated that the proposed DMMF method reduced the total operation cost compared to the operation cost derive from original Cigre 14 bus formulation.