(442e) New Method for Non-Linear Refinery Operations Planning

Typical commercial planning models are linear, usually referred to as "the LP". These models are linear because units inside the refinery are modeled as input-output units with fixed product properties. For example, HDS units establish the output level of sulfur in the exit streams, product fractions of cracking units are usually set beforehand, octane and aromatic levels in reforming are also fixed, etc.

It is very well known, however, that all these reaction units have complex kinetics that depend on temperatures, pressures, space velocities and of course inlet feed properties. One can clearly manipulate these variables within a refinery planning model to obtain a decision making set that leads to a better GRM.

While the above idea is appealing it presents a lot challenges because it leads to large MINLP formulations that are difficult to solve with existing solvers, not even locally sometimes.

To address the computational difficulty we developed a simple linear model that is capable of capturing the nonlinearities of the NLP model effectively. In fact, we believe it can capture global optima efficiently, provided certain smoothness conditions hold. But even without claiming global optimality, our MILP model runs efficiently in reasonable time capturing improved solutions to those of the LP formulation.

To test the formulation, we run a refinery planning problem aiming at the determination of the amount to purchase among six potential crudes that are offered at different prices. The planning horizon is three months and the product demand projection is given. The model includes the hydrogen and fuel balance, two crude units, hds units, reforming, isomerization and blending.

The example solved shows remarkable increase in GRM over the results of the LP model. It accomplishes it by debottlenecking a reforming units manipulation reactor conditions and changing aromatics and octane output in such a way that it allows the processing of additional crude.