Energy management for microgrids using a reinforcement learning algorithm

This paper presents an energy management strategy for microgrids using reinforcement learning. This microgrid is powered by photovoltaic (PV) systems supported by batteries and, in this study, it is intended to be applicable to wireless communication networks. The energy management strategy is designed to address the tradeoff between communication quality of service and energy availability using a learning mechanism that adapts the system to its environment. This learning approach could increase resilience and energy availability of the microgrid. A reward function is designed for base stations so that accurate information of the microgrid is not needed. The learning algorithm was applied and tested in a simulated microgrid, whose performance is comparable to that of a globally exhaustive search.