Font Size: a A A

Global optimization techniques for process systems engineering

Posted on:1999-08-25Degree:Ph.DType:Thesis
University:Princeton UniversityCandidate:Adjiman, Claire Sandrine JacquelineFull Text:PDF
GTID:2460390014471156Subject:Engineering
Abstract/Summary:
New deterministic global optimization algorithms that can be applied to many problems relevant to chemical engineering are presented.; The {dollar}alpha{dollar}BB global optimization algorithm addresses the general class of twice continuously differentiable nonlinear problems (NLPs). This branch-and-bound algorithm is built on rigorous theoretical foundations that enable the convex underestimation of arbitrary twice continuously differentiable functions through the use of interval arithmetic on the Hessian matrices of such functions. Several procedures making use of this information are suggested. The properties of the resulting lower bounding problems guarantee finite {dollar}epsilon{dollar}-convergence to the global optimum solution. The algorithm has been implemented in a user-friendly and automated environment. It includes a number of branching and variable bound update strategies that automatically analyze the structure of the problem to efficiently guide the search of the solution space. A variety of problems have been solved successfully by the algorithm. They include small literature examples, chemical engineering design problems, stability problems, and the identification of all solutions of nonlinear systems of equations.; The {dollar}alpha{dollar}BB algorithm is used in the design of two new global optimization algorithms for nonconvex mixed-integer nonlinear problems (MINLPs). The SMIN-{dollar}alpha{dollar}BB algorithm addresses problems with restricted participation of the binary variables, but general nonconvexities in the continuous variables. It relies on the solution of convex MINLPs or NLPs at each iteration and partitioning in the continuous and binary space. The GMIN-{dollar}alpha{dollar}BB algorithm can solve problems in the much larger class of MINLPs with continuous relaxations involving twice continuously differentiable functions. Branching takes place in the integer domain. Efficient schemes for the generation of valid lower bounds on a partial continuous relaxation of the nonconvex problem have been devised. Both approaches have been tested on several literature examples. Their performance has also been investigated on the synthesis of a heat exchanger network, a pump network and the trim loss minimization problem encountered in the paper industry.
Keywords/Search Tags:Global optimization, Algorithm, Problem, Twice continuously differentiable
Related items