MULTI-OBJECTIVE OPTIMIZATION USING EVOLUTIONARY ALGORITHMS FOR COMPUTATION INTENSIVE APPLICATIONS

Abstract

Evolutionary Algorithm (EA) possesses several characteristics that are desirable to solve real world optimization problems up to a required level of satisfaction. Multiobjective Evolutionary Algorithms (MOEAs) are designed with regard to two common goals, fast and reliable convergence to the Pareto set and a good distribution of solutions along the front. In his work, evolutionary algorithms based approaches for multi-objective optimization have been studied. The particle swarm optimization has been studied in detail. The Particle Swarm Optimization (PSO) is a stochastic, population-based algorithm for search and optimization from a multidimensional space. In this dissertation, multi-objective particle swarm optimization has been implemented for two problem domains. The first problem domain is: designing and optimizing the micro-/millimeter wave components, in this optimal design of two microstrip antenna, (proximity coupled dual-frequency and compact triple-band) and the optimized design of non- linear tapper has been presented. The second problem domain is: association rule mining, in this rules have been generated for two market basket type database (randomly generated and Mondrian foodmart dataset) using multi-objective particle swarm optimization. We have also parallelized the multi-objective particle swarm optimization on GPU for benchmark problem (DTLZ6) and real life problem (association rule mining) and the speed up in running time for both the problems have been presented. The code for optimization is implemented on MATLAB 2006b, and we have used the EM simulator 1E3D for antenna design. The experimental platform used for parallelization is based on Intel(R) Xeon(R) CPU E5420 @ 2.50 GHz, 2.49 GHz, 16.0 GB RAM, NVIDIA Quadro FX 3700, and Windows XP (x64), and we have used visual studio 2005 for the sequential and parallelization code