Improving the robustness and efficiency of crude scheduling algorithms

Higher crude prices have made it even more imperative that refiners blend low-quality and high-quality crudes optimally to maximize their margins. Mathematical modeling of crude blending results in bilinear terms, which combined with commonly used feed quality specifications, make crude scheduling a large, nonconvex, mixed-integer nonlinear optimization problem. The existing literature algorithms and software (DICOPT/GAMS and BARON/GAMS; Brooke et al., GAMS: a user's guide, GAMS, 1998) fail to solve practical instances of this difficult and useful problem. In this paper, we first enhance the practical utility of our previous crude scheduling algorithm by adding 15 properties and corresponding linearly additive indices, which are used in the refinery industry to ensure feed quality. Then, we propose some new iterative strategies to improve the robustness and solution quality of this algorithm. We also propose a partial relaxation strategy to increase its solution speed. We prove its enhanced performance using 24 industry-scale examples, and estimate bounds on solution quality. In contrast to existing algorithms or software that fail to solve most of these problems, our revised algorithm solves all problems successfully and gives profits within 6% of our computed upper bounds. (c) 2007 American Institute of Chemical Engineers.