摘要
利用动量法和自适应改变学习率改进BP神经网络算法的基础上,针对网络权值调整时不容易跳出误差平坦区的问题,进一步对神经网络的学习算法进行了改进,引入一个陡度因子。并把改进算法后的BP神经网络在盐酸浓度的软测量中做了仿真实验,实验结果表明:陡度因子的引入不但可以提高模型的精度而且也使网络的泛化能力得到了增强。
It brought forward importing a factor of gradient to improve the mathematics on the base of the means of momentum and self-adapting modify learning rate, aimed at the training difficult to escape the flat area of error, and carry through a experiment on the process of synthetically hydrochloric acid with the improved BP-neural-network. Experimental results show this method not only can raise precision of the model, but also can enhance the ability of generalization.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2005年第z1期185-186,共2页
Chinese Journal of Scientific Instrument
关键词
BP神经网络
陡度因子
软测量
泛化能力
BP-neural-network Factor of gradien Soft-sensing Ability of generalization
