A Reinforcement Learning based Energy Management System for a PV and Battery Connected Microgrid System

Design of a standalone renewable system to meet the load demand is always complex, as renewable sources are intermittent in nature, which causes overloading on the distribution transformer. However, incorporating a battery system in the design would help in improving the efficiency and mitigate the problem of fluctuating voltages and line loadings. In this paper, a grid-connected PV and battery systems are designed with an objective to manage the energy distribution and meet the load demand. Without the need to know the priori system dynamics, a Q-learning algorithm is used for controlling the battery charge and discharge characteristics (state of charge) based on the load demand and power generated from the PV system. An allocation scheme is developed for the effective usage of the energy sources as well as to increase the life cycle of the battery. Thus, the proposed strategy not only maintains the state of charge of the storage unit but also allocates the usage of the PV source.