Otimização da programação da produção em refinarias para redução de perturbações em unidades de destilação

Abstract

Optimizing the refinery resources is strategic for keeping a company competitive in a fierce and stringent market environment. The main goal of the present work is to develop a formulation for the crude oil scheduling problem addressing operating rules of the Presidente Bernardes Refinery (RPBC) so that its contribution margin is to be maximized. More specifically, load transitions that upset the distillation units is to be tackled so that the tanks containing the different crude mixtures are suitably managed during load exchange. The problem under study involves a set of crude mixture parcels being received through a single pipeline at the refinery tank farm comprised of 11 storage and blending tanks. Up to 36 crude oil types might compose the crude mixtures fed to the three available Crude Distillation Units (CDU). A set of six different scenarios were used to test the formulation performance. The scenarios are distinguished by the composition and the total amount in the initial inventory at the storage and blending tanks, tanks availability, the time at which crude parcels arrive at the refinery as well as their volumes and compositions, and the length of the scheduling horizon. The devised formulation was built taking the MOS model as basis, which uses the continuous time representation. The MOS model has been focus of study by other authors in the past and some changes proposed by those authors that lead to a resulting formulation more adherent to the way Brazilian refineries operate were also incorporated in the formulation proposed in the present work. The innovative contributions of the present research are: modeling of the concept of injecting tanks which are used to complete the CDU load, detailed flowrate control of multiple outputs at storage tanks, and detailed allocation of tanks involved in load transitions at the CDUs, which is the most important cause of operating disturbances of these units. The latter is baptized by Petrobras as the “Polegada” operation. A total of nine formulations were derived as a result of different features incorporated to the MOS base model. The resulting nonconvex MINLP problems were implemented in the GAMS, which demanded the use of tailored solution algorithms. Three strategies were investigated: the mMILP-NLP decomposition, an iterative linear approximation and an out-of-shelf global optimization algorithm. Results demonstrated that the proposed formulation was able to cope with the real-world operations of the refinery under study, generating good solutions from an operational stand point and in reasonable solution times depending on the features incorporated in the model. Modeling of the Polegada operation combined with the concept of injecting tanks enabled smoothening of composition during load exchanges at distillation units, although it caused a considerable increase in solution times and yielding infeasible solutions. Therefore, slack variables were added to the formulation allowing violation of strategic constraints leading to an increase in the number of feasible solutions found, but still demanding computational requirements. As a conclusion, the computational time is hence the most challenging aspect of making such a solution an effective supporting tool for the programmer in real life.