论文部分内容阅读
对遗传算法采用实数编码用于训练神经网络进行了讨论，由实际运算发现，因为只有好的成员才能繁衍，所以无论淘汰率较小或较大时，经过若干代的进化后，往往形成近亲繁衍生息的情况，近亲繁衍时交叉操作对进化作用不大。如果每代只取一个最优成员采用突变方法产生子代成员来训练神经网络能得到很好的结果，最优成员做为子代中的一员时可避免振荡的发生。
Real genetic algorithm is used to train neural network. It is found from the actual operation that only good members can multiply. Therefore, no matter whether the elimination rate is smaller or larger, after several generations of evolution, it often forms the inbreeding Survival of the situation, when inbreeding multiplication of inbreeding little effect on evolution. If each generation takes only one optimal member, using mutation method to generate child members to train neural network can get good results, the optimal member can avoid the occurrence of oscillation as a member of children.