Bilevel programming for steady-state chemical process design—II. Performance study for nondegenerate problems

A simplified model of a process that produces aniline from nitrobenzene is optimized using the active set and relaxation methods developed in Part I. Han's Sequential Quadratic Programming method is used to solve the conventional nonlinear programs for both methods. The performance of each approach is examined for two cases. The chosen starting point for the first case does not require the sequence of iterates to traverse a phase boundary to reach the optimum. In this case, we compare the performance of each of our methods with the other, and with the performance of a more traditional “modular” approach. The results demonstrate that each of the new methods is competitive with the traditional one. The chosen starting point for the second case requires a phase envelope to be crossed to reach the solution. Although each method is able to reach the solution in a comparable number of iterations, the computation time required by the relaxation is nearly 50% more than that for the active set method. This aditional effort is attributed to the nature of the nonlinear programs solved for the relaxation approach.