摘要
进化计算中存在的遗传漂移现象导致种群均匀地收敛于单一个体,产生早熟收敛或丢失可选的全局(或局部)最优解.遗传漂移起源于选择压、选择的采样误差和重组算子的作用.多父代重组算子是多种交叉和重组算子的广义化.以多父代重组算子为对象,研究重组的遗传漂移现象.通过基因频率分析,严格地证明多父代对角交叉和均匀扫描交叉不会引起遗传漂移,但基于基因出现频率的多父代扫描交叉导致遗传漂移,其程度随重组规模的增加而加剧.应用基于基因出现频率的多父代扫描交叉的仿真遗传优化实验的数据表明,重组的遗传漂移不但导致种群多样度的急剧下降,而且降低了进化搜索的收敛速度,从而降低了进化搜索的性能.
Genetic drift in evolutionary computation, which results from the selection pressure, sampling error and recombination, drives the population to converge to a single individual uniformly, thus causing premature stagnation or losing alternative global or local optima. For multi-parent diagonal crossover and scanning crossover are generalizations of the various conventional crossover and recombination operators, genetic drift from which is theoretically analyzed. By analyzing the frequency of the dominant allele, it is strictly proved that multi-parent diagonal crossover and uniform scanning crossover do not cause a genetic drift, but multi-parent occurrence-based scanning crossover induces a strong genetic drift that augments with the increasing of the number of parents. The simulant genetic optimization shows that the genetic drift induced by multi-parent occurrence-based scanning crossover reduces both the population diversity and the convergence rate, thus deteriorating the performance of an evolutionary search.
出处
《软件学报》
EI
CSCD
北大核心
2003年第11期1875-1881,共7页
Journal of Software
基金
国家自然科学基金
湖南省教育厅科学研究基金~~
关键词
进化算法
遗传漂移
重组
Computer simulation
Convergence of numerical methods
Optimization
Theorem proving
