Programação de Blending e distribuição de diesel em refinarias

Abstract

The present work is concerned with the development of alternative optimization models which address the scheduling problem of diesel blending and distribution in oil-refinery operations. Diesel products with different grades are blended considering intermediate streams generated by upstream process units, which are stored in dedicated tanks and are available for use at any point in time along the time horizon. Final products are supposed to be prepared in a manifold (in-line blending) just prior to be sent to the final destination through pipelines, which means that certification time is not imposed in this case, although it is mandatory by law in Brazil. Different diesel parcels are pumped contiguously, in which case an interface between the two adjacent parcels is established creating a certain amount of off-spec material. This fact is addressed in the models by postulating interface identification constraints and penalizing such an occurrence in the objective function with a cost that is usually evaluated based on the diesel products involved with the formed interface. The foregoing mentioned cost is sequence dependent and the addition of interface identification constraints has huge impact on model size and on solution performance. Our study starts by revisiting a model originally proposed in the literature by Pinto et al. (2000). Next, improvements mainly concerning with the interface identification constraints are proposed and evaluated with the aim at expanding model applicability as to the spanned time horizon, which enables the model to be more conformable with real world scenarios. The well known discrete time representation is employed and all proposed models fall into the MILP class of problems which are implemented in the GAMS system and solved with the help of the CPLEX solver. The adopted performance criterion is to minimize costs including raw material, pumping, inventory and transition costs. A comparative analysis of all proposed models was performed. Although the models are somewhat still limited, results look very promising and serve as a basis for future developments in this area where we observe scarce materials in the literature