跨越连接BP算法及其在卷烟制丝工艺中的应用  

BP Algorithm of Neural Networks with Cross Connections and Its Application in Cigarette Primary Processing

作　　者：唐云岚[1] 高妍方[1] 赵青松[1] 陈英武[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院

出　　处：《系统仿真学报》2008年第20期5523-5525,5529,共4页Journal of System Simulation

基　　金：国家自然科学基金资助项目(70272002)

摘　　要：传统BP神经网络算法及其改进算法都是非完全全连接神经网络算法,具有收敛速度慢、泛化能力差等不足。通过对神经网络连接方式的转化,可以得到一种完全全连接神经网络——跨越连接神经网络,并给出了跨越连接BP神经网络算法。针对卷烟制丝工艺的仿真实验表明,该算法具备描述复杂数据的能力,与传统BP算法相比,网络训练收敛速度快且泛化能力强。Traditional BP algorithm and improved BP algorithms are kinds of half-fully connected neural network,which have the problems of slow astringency and poor generalization ability. By transforming chaining mode of the neural network,the completely-fully connected neural network model-neural networks with cross connections could be got. Further more,the new BP algorithm of neural networks with cross connections was proposed. According to problems lying in cigarette primary processing,experimental results show that the new algorithm has the ability of describing complex data. Compared with traditional BP algorithm,this new method has the characteristics of fast astringency and better generalization.

关 键 词：完全全连接神经网络 跨越连接 BP算法 卷烟制丝工艺 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]