A Coupled Strategy for the Solution of NLP and MINLP Optimization Problems: Benefits and Pitfalls

Abstract

This paper presents a strategy for the solution of nonlinear programming (NLP) and mixed-integer nonlinear programming (MINLP) problems, based on the coupling of the equation-oriented simulator ASCEND IV with the stochastic optimizers MSGA and MSIMPSA. Both NLP and MINLP formulations of a reactive distillation example were explored. The results show that the connection between the two software blocks was successfully established. Despite the highly nonlinear behavior of the proposed example, the results presented generally agree with those previously obtained for the same case study using other approaches, namely, SIMOP, which is a FORTRAN 77-based NLP simulation tool developed for optimization problems. Increasing the column dimension or switching the vapor−liquid equilibrium (VLE) description to a nonideal situation, which produces nonlinear models with a much larger number of equations that must be solved simultaneously, show that the results deteriorate. This is attributed to the lack of user control over the subsystems’ structures (viz, tearing variables). This problem was partially circumvented through the application of a more elaborate scheme initialization for the column variables.