Sizing Grid-Connected Microgrids Based on Deep Reinforcement Learning

Nowadays, renewable energy resources have been widely installed in power system around the world. However, how to size and operate the grid-connected microgrids to achieve cost-effective is an essential problem. The problem is a high dimension nonlinear optimization problem, and the existed methods cannot meet all objectives. Deep reinforcement learning (DRL) is a model-free approach and has the advantages to deal with complex nonlinear model, in addition, it can deal with uncertainties by large amounts of data training. Thus, DRL is suitable to solve the sizing optimization problem. In this article, we present a DRL based sizing approach for grid-connected microgrids. First, the sizing problem is transferred into a step by step decision problem; second, the DRL based solving algorithm is developed, including the states, actions, and rewards; last, deep deterministic policy gradient (DDPG) algorithm is implemented to search the best sizing results. The simulation results reveal that DRL has good performance to solve the sizing problem, and provide a novel idea to solve the combinational optimization problem.