遗传算法与神经网络(Ⅱ)──用EGA-GDR训练神经网络

GENETIC ALGORITHMS AND ARTIFICIAL NEURAL NETWORK(Ⅱ)──TRAINING ARTIFICIAL NEURAL NETWORK BY EGA-GDR

为了更有效地解决神经网络培训中遇到的局部最优解问题，提出一种融合EGA和GDR算法特长的新算法──EGA-GDR算法．通过分别对拓扑为2×3×3和8×8×4两个神经网络的训练，表明该算法较之SA-GDR算法可以更有效地解决神经网络训练中的局部最优解问题．

In order to overcome a local optimal solution which often occurs in the training process of ANN,a new algorithm,called EGA-GDR,is proposed,which combines the perfect characters of both EGA and GDR.The results obtained from training two ANNs whose structures are2×3×1 and8×8×4 respectively,show that EGA-GDR can more effectively search the optimal solution than SA—GDR does.