改善遗传神经网络性能的研究

A study on improving the performance of genetic neural networks

本文设计了一种基于非近亲配对的遗传算法,该算法在对当前种群进行遗传选择操作时,选用了非近亲配对的选择方法,从而避免了个体单一导致早熟的现象。并将该算法用来优化前馈神经网络的连接权值。通过对异或问题的前馈神经网络的性能测试,证明了基于非近亲配对遗传算法的神经网络的性能是优于基于传统的比例选择遗传算法的神经网络的。

a new genetic algorithm based on the irrelative mating was proposed. The premature convergence was avoided for the irrelative mating. Applied to the problem of optimizing the connection weights of the feed-forward neural networks, the algorithm was feasible. Through a group of tests to the performance of feed-forward neural networks such as XOR, it was proved that the irrelative mating genetic algorithm was better than genetic algorithm based on proportional selection in practice.