Economic dispatch of industrial park considering uncertainty of renewable energy based on a deep reinforcement learning approach

Abstract

The proportions of renewable generations (RGs) and flexible loads (FLs) are rapidly increasing in the industrial park. However, the randomness of RGs and FLs challenges the economics and complexity of the industrial park dispatch. To take this challenge, this paper proposes an optimal dispatch strategy for the industrial park based on a deep reinforcement learning (DRL) approach. Firstly, RGs are modeled based on random fuzzy theory. Dynamic line rating (DLR) is introduced to improve the power flow capacity of RGs model in transmission lines. Secondly, in dispatch process is modeled as a Markov decision process (MDP) to avoid complicated linearization processes. Thirdly, we propose a variant of the distributed proximal policy optimization (DPPO) algorithm. The distributed agents in the algorithm are simultaneously trained to explore optimal strategies in different environments. The trained agents can complete the dispatch based on the current state of the industrial park. The effectiveness of the proposed method is verified by numerous simulations. Compared with PPO, DDPG, and TD3 algorithms, the training time of the proposed approach is reduced by 64.06%, 47.31% and 46.99%, respectively. In different scenes, the operating cost with the proposed method is reduced by 47.19%, 46.23%, 43.41% and 41.71%.