Inventory Pinch Based Multi-Scale Model for Refinery Production Planning

In this work, we introduce a new two-level decomposition algorithm for production planning. The top level optimizes operating conditions via a nonlinear (NLP) model, while the lower level computes an optimal production plan by solving a mixed-integer linear (MILP) model which uses operating states computed at the top level. Time periods at the top level are delineated by inventory pinch points. The algorithm solves mixed-integer nonlinear (MINLP) models much faster than current MINLP solvers for our case studies associated with gasoline blend planning and refinery production planning.