基于PCA-改进算法BP神经网络的软测量技术  

Soft-sensing technology based on PCA-improved BP neural network

作　　者：曹先庆[1] 

机构地区：[1]沈阳化工学院,沈阳110142

出　　处：《仪器仪表学报》2006年第z3期2342-2343,共2页Chinese Journal of Scientific Instrument

摘　　要：本文在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。并与主元分析相结合,形成了PCA-改进算法的BP神经网络。通过在塑料缠绕过程塑料张力预测中的仿真实验结果表明:PCA-改进算法后的BP神经网络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。This paper brought forward 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. Considering the influence of the correlativity of input variables, this paper inducted PCA algorithm.ad carry through a experiment on the process of the plastics tension prediction with the PCA-improved BP neural network, simulation results show:this method not only can raise precision of the model, but also can enhance the ability of generalization.

关 键 词：BP神经网络 陡度因子 主元分析 泛化能力 

分 类 号：TH7-55[机械工程—仪器科学与技术]