ITR: Entropy Based Multi-Objective Genetic Algorithm With Constraint Handling and Set Quality Metrics

University Of Maryland, College Park, College Park MD

Investigators

Abstract

This Information Technology Research project aims at developing an entropy based multi-objective genetic algorithm for design optimization. The algorithm is based on an analogy from statistical theory of gases in order to obtain a fullest possible representation of solutions. In the proposed algorithm, statistical behaviors of a sample of designs in a population will be simulated, as they evolve, according to those of the molecules of an ideal gas undergoing expansion in an enclosure. The goal of this analogy is to obtain solutions with maximum entropy, that is, maximum possible uniformity and coverage along the Pareto frontier. The investigation will involve the development of new methods for handling constraints in multi-objective design optimization as well as new quality metrics for assessing the "goodness" of a set of equally optimum (Pareto) solutions. The quality metrics will also be used to develop new convergence criteria for the algorithm. If successful, the outcome of the research will advance the state of the art at the level of computational tools that enable multi-objective engineering design optimization with applications to a wide class of problems. The results of this investigation will be broadly disseminated for the engineering design automation community, transferred and integrated into several courses at the graduate and undergraduate levels at the University of Maryland, and transferred to industry.