自适应小生态遗传算法的理论分析和加速技术

提出了联赛选择和相似个体概率替换的自适应小生态遗传算法，建立了小生态生长的动力学模型．平衡态理论分析和仿其实验表明，概率联赛小生态技术选择能够形成和维持稳定的子种群．提出了种群聚类分割和单纯形搜索的并行局部搜索算于，定性地分析了其搜索性能．对复杂多峰问题的优化结果表明，结合概率联赛选择和并行局部搜索算子的小生态遗传算法不但能够快速可靠地收敛到全局最优解，且能并行地搜索到多个局部最优解，其收敛速度和全局收敛可靠性均显著地优于简单遗传算法和其它小生态方法．

Theoretic Analysis and Accelerating of a Class of Self-Adaptive Niching Genetic Algorithms

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.