Lagrangean Decomposition for Integrated Refinery-Petrochemical Short-term Planning

We present a methodology for the optimal integration of crude management (CM) and refinery-petrochemical (RP) planning operations. The physical coupling between both CM and RP optimization subproblems is via the flow rate, physical-chemical properties, and composition of the crude blends. For a given economic cost of the crude blends, which either provides a selling price for CM or a purchase price for RP, both subproblems can maximize their profits independently. But failure to integrate these two subproblems can create an imbalance between crude supply and demand. Optimizing CM and RP operations simultaneously entails the solution of large-scale, nonconvex quadratically-constrained quadratic programs (MIQCQPs). We apply a spatial Lagrangean decomposition algorithm to tackle these MIQCQPs and demonstrate it on a full-scale industrial facility. The results show that Lagrangean decomposition can outperform commercial global solvers BARON and ANTIGONE when applied to the monolithic MIQCQP. The Lagrangean decomposition can also reduce the optimality gap faster than with a clustering decomposition algorithm, leading to optimality gaps below 5% within 1 hour of CPU time.