Abstract: Focusing on the application of deep reinforcement learning in crude oil scheduling, the scheduling process is modeled as a Markov decision process, and the Dueling DQN algorithm is adopted to solve the crude oil scheduling optimization problem. To address challenges such as disparate state scales and fluctuating reward distributions, state normalization and reward standardization mechanisms are designed to enhance training stability and convergence efficiency. Furthermore, the complexity of the action space is effectively reduced through the strategic simplification of unloading decision dimensions. Experimental results demonstrate that the proposed method can generate stable, feasible, and cost-effective scheduling plans while satisfying various process and operational constraints. These outcomes verify the effectiveness and application potential of deep reinforcement learning in optimizing complex crude oil scheduling tasks in refinery operations.

Key words: crude oil scheduling, deep reinforcement learning, Dueling DQN, scheduling optimization