Theoretic Analysis and Accelerating of a Class of Self-Adaptive Niching Genetic Algorithms
Guo Guan · Chinese Journal of Computers · 2003
This paper proposes a kind of self-adaptive niching genetic algorithm (NGA) using probabilistic tournament selection. NGA likely accepts the winner of a parent and an offspring with similarity as a member of the next population. The dynamic equation of the niche proportion is formulated by expectation proportion analysis. The analytical solution in equilibrium for two niche problem proves that NGA is capable of forming and maintaining stable subpopulations, which is verified by experiments. This paper also proposes a parallel local search operator (PLS) that implements clustering partition of the population and simplex local search. PLS divides the population into a group of disjoint subpopulations, each of which consists of several individuals with neighboring space locations. It performs independent local search within each subpopulation by simplex method. The reliable global exploration of NGA and fast local convergence of PLS within niches not only locate various local optima concurrently,but also increase the convergence speed remarkably. The experimental results optimizing various classes of test functions show that, NGA+PLS is a much more competent optimization method than canonical genetic algorithms and other niche methods.