(236a) A Novel Strategy for Global Optimization of MINLP Models

Chemical engineering problems are often modeled as of non-linear models, which many times are also non-convex. Global solutions are frequently required, especially when important decisions need to be made. Although in some cases there is no strict need of finding the global optimum solution (gap zero), it is very important to have at least an idea of how much better a found solution could be.

In this paper we present a global optimization method that relies on the partition of some critical variables (or functions) that are strategically used as reference. We replace the non-convex non-linear constraints with certain discrete bounds to generate lower bounds.

The lower bound model is also used to eliminate portions of the feasible region that cannot be part of the global optimum solution. The elimination procedures have the advantage of eliminating portions of the feasible region based on solution feasibility between lower and upper bound. In other words, there is no need of finding the global optimum lower bounds at each step of the elimination procedure.

Additionally, when the problem is extremely degenerate, the critical variables can be partitioned and two (or multiple) sub-problems are analyzed. This second recourse is a branch and bound procedure.

Examples will be shown.