Reinforcement Learning Based Energy Internet System Optimal Dispatch: A Survey

The energy internet can meet diversified load-side energy demands, improve energy utilization and promote sustainable energy development through the complementary synergy of electricity, gas, cooling and heat networks. However, the multi-timescale coordination of multiple coupled heterogeneous energy sources, the multi-subject game interaction of source-grid-load-storage, and the high-dimensional non-convexity of the optimization calculation all lead to many challenges for optimal dispatch of the energy internet. As a topical trial-and-error learning method, reinforcement learning is suitable for solving complex dynamic system optimization with uncertainty, therefore, it is gaining widespread interest in optimal dispatch of the energy internet. This paper reviews researches on reinforcement learning-based solutions to optimal dispatch of the energy internet at the model and algorithm, and presents perspectives on both multi-timescale optimization and transfer learning.