Quadratic approximation based hybrid genetic algorithm for function optimization

Abstract

Probably the popular form of binary genetic algorithms for function optimization use tournament selection (TS) or roulette wheel selection (RS) for function optimization. Also single point crossover (SC) and uniform crossover (UC) are most popular and effective crossover operators. In an earlier paper we had considered all four combinations of these crossover and mutation operators along with bit-wise mutation, called GA1 (TS + SC), GA2 (TS + UC), GA3 (RS + SC) and GA4 (RS + UC). In this paper, an attempt is made to hybridize these four GAs by incorporating the quadratic approximation (QA) operator into them. The four resultant hybrid GAs, called HGA1, HGA2, HGA3 and HGA4, are compared with the four simple GAs on a set of 22 test problems taken from literature. Based on the extensive numerical and graphical analysis of results it is concluded that the HGA3 outperforms all rest 7 versions. Further, we study the depth and frequency of the QA should be applied for better performance for the particular problem suite.