Optimal Refinery Crude Oil Scheduling Based on Proximal Policy Optimization

Currently, most refinery crude oil scheduling studies adopt static scheduling schemes based on mathematical programming, which cannot adjust and optimize according to environmental change in real-time. This paper establishes a dynamic real-time scheduling decision model subject to refinery production constraints and designs the corresponding agent interaction environment. The proximal policy optimization (PPO) algorithm in deep reinforcement learning solves the model. The performance of the strategies learned in each production link is evaluated and analyzed through experiments, and the adaptability of the model is evaluated under the uncertain condition of the arrival time of the tanker. The results demonstrate that this model can obtain scheduling results compatible with business evaluation, thus effectively reducing the influence of random events on the overall decision and improving the performance and stability of crude oil scheduling.