INTEGRATION OF LOGIC AND HEURISTIC KNOWLEDGE IN MINLP OPTIMIZATION FOR PROCESS SYNTHESIS

This paper proposes a quantitative framework for the integration of logic and heuristic knowledge that is expressible in propositional logic form in MINLP optimization models for process synthesis. The objective is to use this type of qualitative knowledge to expedite the search, but without compromising optimality of the solution. The basic idea relies on converting logic relations among units in a superstructure and heuristic design rules into a set of linear inequalities. Having obtained such a model, strategies are proposed for its integration within mixed-integer nonlinear programming (MINLP) techniques at the levels of model formulation, and algorithmic search for the Generalized Benders Decomposition and Outer Approximation methods. Basic properties of the formulation are given, as well as a systematic method for adjusting weights for violation of heuristics. The application of the proposed method is illustrated with several process synthesis problems to show that improved computational efficiency and robustness can be achieved.