Optimal Scheduling of Integrated Energy System Based on Multi-agent Reinforcement Learning

The different interests pursued by each entity and the complex coupling relationship between multiple energy sources pose many challenges to the optimal operation of multi-park integrated energy systems(IESs). To this end, this paper proposes a data-driven based multi-agent proximal policy optimization (MAPPO) algorithm for the optimal scheduling of multi-park IESs. Firstly, the real-time optimal scheduling problem of multi-park IESs considering energy trading and carbon market trading is modeled as a Markov decision process (MDP). Secondly, the joint evaluation of MAPPO centralized training decentralized execution(CTDE) framework and global value function is used to achieve distributed autonomous learning for each agent to improve the quality of scheduling policies and privacy security of each park. The simulation results show that the proposed method has a faster convergence speed and can reduce the operating cost of the system under the influence of random fluctuations of new energy output and load.