Optimal Dispatch Strategy of Integrated Energy System Based on Deep Reinforcement Learning Considering Security Constraints

Under the "dual carbon" goals, the couplings and cascade utilizations of the multi-energy systems in an integrated energy system have become an important measure to achieve the goals. As a complex, neither linear nor convex problem, the optimal operation of an integrated energy system has resulted in difficulties in obtaining a global optimal dispatch strategy in the traditional methods. Meanwhile, the problem is further compounded by the increasing penetration of renewable energy sources, such as solar and wind power, with the growing complexity of the network topology. The reinforcement learning provides us with an effective way to solve the above problem, however, most of the researches on the reinforcement learning-based optimal dispatch seldom consider the overall security constraints at present. Therefore, a deep reinforcement learning-based optimal dispatch model of the integrated energy system considering the security constraints is proposed in this paper. Firstly, the energy flow constraints of the power and heat are considered in the optimal dispatch model to realize the security verification of the dispatch strategy. Secondly, the deep reinforcement learning algorithm is implemented to convert the conventional dispatch problem into a reinforcement learning sequential-decision problem, greatly improving the design of the state spaces, the action spaces and the reward function of the agents. Finally, the optimal dispatch decisions are solved and applied in the offline and online environments. The effectiveness and rationality of the proposed method are verified by comparison and analysis of case studies. © 2023 Power System Technology Press. All rights reserved.