Improvement of Crude Oil Refinery Gross Margin using a NLP Model of a Crude Distillation Unit System

Abstract This work presents a Non Linear Programming (NLP) model developed to optimize simultaneously a crude oil distillation unit (CDU) system and several cases of application run in a refinery as well. This model optimizes feedstock composition and operational conditions for a CDU System (ECOPETROL S.A.). The NLP Model uses a Metamodeling approach so as to represent Atmospheric Distillation Towers (ADT). The Vacuum Distillation Towers (VDT) are implemented assuming perfect separation (assay cuttings). The defined objective function is given by an economic profit. The CDU system consists basically of five industrial units and fourteen Colombian Crude Oils. Each Metamodel uses as independent variables: crude oil flow rates, operational conditions, Jet EBP, and Diesel T95% from ASTM D-86 distillation curve. The output variables of the Metamodels are product flows, temperatures, and qualities. The developed NLP model was implemented in GAMS. The time needed for its solution is around 60s while using the CONOPT solver. The NLP model results were successfully applied to a Colombian refinery for 3 consecutive weeks. The model was able to find the best use of installed equipments in CDUs through the preparation of a crude oil charge quasi-constant quality without matter the time period of the optimization. In each week, optimal crude oil flow rates towards each CDU (like new scenarios implemented in the refinery) were evaluated in a refinery global simulator with all downstream refining schemes in order to calculate the Refinery Gross Margin (RGM). In each analyzed case, the obtained RGM for new crude oil feeds was however better than that case without optimization with a economic benefit of up to 0.043 US$/bl equivalent to US$ 3.870.000 per year. This shows the effectiveness of a CDU NLP model within short term planning in the petroleum industry.