Multiperiod design optimization with SQP decomposition

Abstract

This paper presents a new decomposition method for solving large scale multiperiod design problems. These problems are formulated as nonlinear optimization problems with a special block angular structure and linking variables. The proposed method, based on successive quadratic programming (SQP), uses a decoupling scheme and projects the orignal problem into a quadratic subproblem involving only the complicating variables. Using this subproblem as the main coordination step the problem in the full space is solved as a stream of independent single period problems. The key property of this method is that the computational effort scales linearly to the number of periods compared to a quadratic and cubic increase for general purpose reduced gradient and SQP methods, repectively. To illustrate this property, the method is applied to four example problems including two in multiperiod chemical process design. Its performance is superior to both MINOS and SQP in terms of computational demands, number of function evaluations and solution robustness.