ABSTRACT

ABST R AC T Process synthesis through superstructure optimization techniques are generally regarded as theoretically powerful; however, they have not been widely used in practice since they typically result in large-scale nonconvex mixed-integer nonlinear programming (MINLP) models which can not be solved effectively. To address this limitation, we propose a framework leading to substantially simpler formulations which can be solved to optimality. This new approach includes the replacement of complex first-principle unit models by compact and yet accurate surrogate models. We show how all the relevant variable relationships established by a unit model, can be expressed in terms of a subset of the original model variable set. We discuss how this subset of variables can be identied, and we present a method to develop high quality surrogate models through articial neural networks. Finally, a superstructure optimization problem is presented to illustrate the proposed framework.