摘要
针对误差反向传播(BP)算法和遗传算法各自的优点和不足,提出了遗传算法优化神经网络技术:利用遗传算法的全局搜索能力,对神经网络连接权进行优化,以遗传算法优化的初值作为BP神经网络的初始权值,再用BP算法训练网络.优化后的BP网络其误差的递减速度和收敛速度都比标准BP网络快,而且对学习速率调整要求更少.将遗传神经网络应用于混合气体定量识别的训练中,得到的最大误差由20.7%降为12.1%,平均误差从5.4%降为3.5%,识别效果得到了提高.
A genetic algorithm optimizing neural network （GA-NN） is given, after genetic algorithm and back-propagation （BP） neural network were studied. Optimizing the weights of neural network with the character of local search ability of genetic algorithin, the optimized value was used as the initial weights of the back-propagation neural network, and then the network was trained by the backpropagation method. The results show that the convergence speed and precision of genetic algorithm optimizing neural network are better than that of the single algorithm. The application of genetic algorithm optimizing neural network to the recognition of multi-gas validates that the method improved the detection effect of multi-gas with reducing the maximal error and average error from 20. 7 % and 5.4 % to 12.1 % and 3.5 %.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2007年第9期118-120,128,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
武汉市科学技术局资助项目(20041003068-02)
关键词
遗传算法
BP神经网络
遗传神经网络
气体识别
genetic algorithm
back-propagation neural network
genetic neural network
gas recognition
