An algorithm development for solving inexact simulation problems

Abstract

Simulation with inexact constraints can be formulated as an optimization problem, characterized by many variables and constraints, but relatively few degrees of freedom. The solution can be obtained by an SQP algorithm whose Hessian can be updated in the space of the design variables alone. Some difficulties are met in solving problems with a high number of design variables and/or with highly non-linear models. Several strategies have been suggested to overcome those difficulties. To test these and others strategies we have formulated a very flexible algorithm where each method can be activated or not. By a suitable test design on two Enichem processes we have observed: (1) a range and null-space projection algorithm with orthogonal bases is useless to improve problem solution; (2) a feasible path algorithm and analytical Hessian approximation significantly improve problem solution. We are programming a larger test plan to analyse better the behaviour of our algorithm on several flowsheeting and simulation problems.