A Deep Reinforcement Learning Approach for Refinery-wide Scheduling

Due to the large production scale, numerous facilities, and tightly integrated processes in refinery-wide optimization, model-based methods face practical challenges when applied across an entire plant. Therefore, this study adopts a data-driven approach, leveraging hierarchical learning of optimal strategies to directly guide production scheduling decisions, offering a novel perspective and solution for refinery-wide scheduling. This paper proposes a hierarchical scheduling optimization method for refinery-wide scheduling based on deep reinforcement learning (RHDL). The upper-level comprehensive scheduling is trained and solved using deep reinforcement learning, while the lower-level sub-models — crude oil scheduling, unit operation scheduling, and oil blending scheduling — are addressed through mathematical programming techniques. This integrated approach enables optimized scheduling across the entire refining process.