Global Optimisation of Gasoline Pool Blending Using Constraint Partitioning

Abstract

Aims: A hybrid Nonlinear Programming–Simulated Annealing method has been applied to solving the constrained offline gasoline recipe optimisation problem using constraint partitioning. Methodology: The method was demonstrated by applying it to a small blending case study with eighteen independent variables where one of the variables was used as a link variable between the two sub-problems of the partitioned non-convex problem. It is noted that this can in theory be extended to larger tightly constrained problems with more link variables e.g. whole refineries where the models involve huge numbers of nonlinear equations and many process units. Results: The approach exhibited good performance representing significant savings against both a derivative-based NLP method used alone and a Mixed Integer Non-Linear Programming method. This performance was examined by way of a sensitivity analysis of the simulated annealing parameters. Conclusion: The convergence times were in minutes and are realistic for short-term recipe optimisation.