Reinforcement Learning Based Energy Scheduling for Pelagic Islands

Pelagic islands (PI) are rich in renewable energy, but due to the natural geographic isolation between the islands, the PI show a pattern of reverse source-load distribution. In this paper, an energy scheduling model based on electric vessels (EV) is designed, and use the reinforcement learning algorithm for agent continuous interaction with the environment, learn the optimal action decision. Firstly, according to the characteristics of the renewable energy endowment of the PI, it is more feasible to use the abundant renewable energy power supply of the islands than the traditional power supply. Further, in view of the differentiation of resources between resource enrichment islands (REI) and habitat islands (HI), the use of EV as power transmission tools for inter-island energy transmission is proposed. On this basis, the energy flow transmission process between PI is described as Markov decision process, the PI energy transmission model is designed in detail using reinforcement learning methods. Consider the PI as an agent, design the state space, action space and reward or punishment of the agent, and then customize the PI environment according to the API of the Gym environment. Finally, PPO algorithm in stable_baselines3 is used to explore the agent's behavior strategy. The experimental results show that compared with the traditional optimization and control methods, reinforcement learning without relying on the environment model can effectively explore the energy scheduling problem of PI.