Control Strategy of Microgrid Energy Storage System Based on Deep Reinforcement Learning

As a new form of energy management, a microgrid has developed rapidly in recent years. In order to ensure the safe, stable and economical operation of a microgrid system, it is important to provide it with reasonable energy scheduling strategy. According to their operating modes, micro-grids can be divided into two categories: grid-connected microgrids and island microgrids. This paper takes the grid-connected microgrids as the object, applies the Simulink simulation technology to build a microgrid system including an external power supply, photovoltaic power generation, energy storage, and load according to the principle of constant power control (PQ control). Based on this simulation system, this microgrid system is combined with the double deep Q network algorithmfirstly. Then it is trained to get an optimization strategy of energy storage control problem, the goal of which is to minimize the 24-hour electricity cost while meeting the voltage deviation of the microgrid, power balance and the constraint of state of charge of the energy storage. Through experimental verification, the rationality of the energy storage strategy is analyzed from a qualitative perspective, and the effectiveness of the method proposed in this paper is demonstrated from a quantitative perspective. © 2021, Power System Technology Press. All right reserved.