Crude Oil Scheduling for Coastal Refineries with Long-Distance Pipelines: Application of Mixed-Integer Programming and Supervised Learning

Abstract

The transportation of crude oil in coastal refineries via long-distance pipelines is a crucial step in refinery scheduling plans. However, existing studies oversimplify the issue by assuming either instantaneous transmission of crude oil or fixed transportation times in long-distance pipelines, disregarding the flow rate fluctuations of crude oil in these pipelines. This oversimplification fails to capture significant transport delays and crude holdups, which can significantly deteriorate the operations in coastal refineries. To address this issue, we study long-distance pipeline transportation under a discrete-time model. We propose a mixed-integer programming model which can accurately describe the nonuniform speed transportation process, and effectively handle refinery scheduling problems involving long-distance pipelines. In addition, we employ a supervised learning method to construct an offline predictor which can reduce the online solution time by minimizing the combinatorial search among discrete variables. In our numerical experiments, we illustrate the proposed model using several real-world coastal refineries as examples. The results show that the model can accurately describe the realistic transportation characteristics of long-distance pipelines, and the generated scheduling scheme can avoid frequent pipeline switching in storage tanks, which can eventually lead to an enhancement of overall refinery performance.