Online and Offline Approximations for Population based Multi-objective Optimization

摘要: Title of Dissertation Online and Offline Approximations for Population Based Multi-Objective Optimization Genzi Li, Doctor Philosophy, 2007 directed by: Professor Shapour Azarm Department Mechanical Engineering University Maryland The high computational cost population based optimization methods has been preventing applications these to realistic engineering design problems. main challenge is devise approaches that can significantly reduce the number function (or simulation) calls required in such methods. This dissertation presents some new online offline approximation optimization. In particular, it DOE metamodeling techniques Genetic Algorithm (GA) multi-objective along four research thrusts. first thrust called: Metamodeling Assisted Fitness Evaluation. this thrust, a assisted fitness evaluation approach developed aims at reducing each generation MultiObjective (MOGA) second Metamodeling. introduces method generations MOGA. It shown under can, compared further third presented sampling points non-smooth regions space order improve accuracy metamodel. useful when available limited. Finally, fourth Dependent MultiResponse Simulations. technique an simulation multiple responses. Numerous numerical examples are used demonstrate applicability performance proposed techniques. situations where application requires numerous evaluations calls), approach, be employed construct global model calls. Moreover, simulations with responses, dependent reasonably accurate metamodels thus facilitate ONLINE AND OFFLINE APPROXIMATIONS FOR POPULATION BASED MULTI-OBJECTIVE OPTIMIZATION