基于动态比例变换的高效遗传算法

HIGH PERFORMANCE ACHIEVED IN GA USING DYNAMIC LINEAR TRANSFORM

遗传算法中,分别以在、离线性能表征算法的进行性能和收敛性能[1],两者是一对矛盾体.在算法进行的早期,人们希望算法有较好的在线性能,以便能快速地搜索到最优点的附近.否则便会出现算法过早收敛的情况,谓之“早熟”.在群体演化的后期,很可能会出现下列情况:个体适应度之间的差值比群体最小适应度相差若干个数量级,此时群体演化的速度会非常慢,甚至很多代也不会达到最优点. 遗传算法的作用原理用模式理论能得到很好的解释.根据模式理论,群体中第j个个体通常以概率pj=fj/∑fj的概率被选择复制.若包含于群体中的某模式H在当前代中有M（H,T）个代表个体。

The dynamic linear transform (DLT) of fitness function is presented to cope with problems in genetic algorithm (GA), which always converges quickly in the early stage and very slowly in the later stage. How to choose parameters for DLT and how it works are analyzed. A hybrid algorithm called Optimal Point Searching by Interpolation is presented through analyzing the neighboring area. The later method can pick out the optimal points using the information available. Examples show that these methods work well.