A data-driven approach for crude oil scheduling optimization under product yield uncertainty

Abstract

Crude oil scheduling plays an important role in reducing production cost of refineries. However, fluctuations in operation conditions in crude distillation units (CDUs) and quality of crude oil will lead to uncertainties of product yields that may nullify the plan obtained from the deterministic model. A data-driven robust optimization (DDRO) method for crude oil scheduling is proposed in this study to address these uncertainties. Historical data of the product yield of crude oil are fully utilized by a generalized intersection kernel support vector clustering (GIKSVC) algorithm to construct uncertainty sets. A new DDRO model is then developed on the basis of derived uncertainty sets and further reformed as a solvable mixed integer nonlinear programming (MINLP) problem via dual transformation. Case studies from a refinery are performed to indicate the performance of this method and the influence of the regularization parameter ν on the optimization solution is explored to reach an acceptable balance between total production cost and robustness of crude oil scheduling.