Evoultionary algorithms are search processes, inspired by concepts from evolution, which find good solutions to hard problems in a reasonable time. Though EAs are computationally simple, they are powerful and robust in searching complex spaces, and are not limited by assumptions about the search space.
But, real applications such as VLSI layout problems sometimes require a large number of individual and many generations. In such case, parallelization of EAs is a promising approach to find beeter solutions in shoter times. PEA performance is mainly affected by migration method, and many researches on PEA are about migration scheme.
This paper presents two migration schemes. They are selective migration method and environmnet variable migration. First method promotes diversity of algorithm, and second method provides higher convergence environment. To test the performance of proposed methods, six unconstrained problems and five constrained problems are evaluated for global minima.
Results indicate that a significant increase in performance can be had by implementing PEA using proposed migration mathod. And, the applicability of proposed methods for real problem is demonstrated by neural network structure optimzation.
