Dealing with Multiple Tank Outflows and In-Line Blending in Continuous-Time Crude Oil Scheduling Problems

• Published in: Industrial & engineering chemistry research Vol. 58; no. 11; pp. 4495 - 4510
• Main Authors: Neiro, Sergio Mauro da Silva, Murata, Valéria Viana, Jahn, Bárbara, Roland Seixas, Raiana, Horn Hollmann, Estefane, Salgado Pereira, Cristiane
• Format: Journal Article
• Language: English
• Published: American Chemical Society 20.03.2019
• ISSN: 0888-5885; 1520-5045
• DOI: 10.1021/acs.iecr.8b03749

The crude oil scheduling problem has been focus of many studies in the past, which is justified by its importance in the oil industry. Optimizing crude oil blending is of paramount importance given that it can substantially impact the economic performance of a refinery. In this work, operational features of a real-world existing refinery are addressed. Only operations limited to the refinery battery are in scope such as splitting of parcels unloaded from a supplying pipeline segment, tank heels and capacities, brine settling time, multiquality tracking, crude distillation unit (CDU) straight-run products profile, multiple tank outputs, and multiple CDU inputs. Moreover, different policies as to the handling of the refinery tank farm are evaluated. The presented formulation is taking from the multioperation sequencing (MOS) proposed by Mouret, Grossmann, and Pestiaux ( Comput. Chem. Eng. 2011, 35, 1038−1063. DOI: 10.1016/j.compchemeng.2010.07.007 ). The novelty of the present work relies on how to accurately impose lower and upper bounds on the flow rate of multiple tank outputs given that the formulation is based on a unit-specific time grid. Two approaches are proposed and evaluated. The resulting MINLP models are exhaustively tested with six distinguishing real-world scenarios in which blending can include up to 36 crude oil grades, different tank availability, initial inventories, and scheduled parcels to arrive at the refinery. Two solution algorithms that avoid solving the full-scale MINLP problem are used and compared. The computational experiments show that the proposed formulations are able to handle a wide range of problem instances in reasonable computational time, despite of the size dimension.