Efficient Virtual Power Plant Dispatch Model for Distributed Energy Resources Using Deep Reinforcement Learning

In the summer of 2022, the southwestern region of China experienced the longest-lasting extreme heat climate since 1961. During this period, hydropower, which serves as a significant energy source, suffered from a severe water shortage, while residents faced a sharp increase in cooling demand, resulting in a significant power shortage. As a modern load aggregation and regulation technology, virtual power plants (VPP) can effectively utilize distributed photovoltaic resources and implement demand response strategies to adjust loads. This paper proposes a VPP scheduling method based on the deep deterministic policy gradient (DDPG) reinforcement learning algorithm and conducts a case study based on the real data from City X in southwestern China. Test results indicate that the proposed DDPG-based VPP scheduling method in this paper has higher training efficiency compared to other similar algorithms, and it can achieve the intelligent control of air conditioners through Smart Home device. The comprehensive cost under this dispatch method has only a 2.03% gap with the theoretical optimum.