神经网络活化函数的研究及其应用  

RESEARCH ON THE ACTIVITY FUNCTION IN ANN AND ITS APPLICATION

作　　者：潘丹[1,2] 罗干英[1,2] 黄茜[1,2] 肖诗铁[1,2] 

机构地区：[1]华南理工大学电子与通信工程系 [2]华南理工大学应用化学系

出　　处：《华南理工大学学报（自然科学版）》1998年第4期49-54,共6页Journal of South China University of Technology(Natural Science Edition)

摘　　要：通过构造一类新的高效分段活化函数，很好地解决了ＢＰ算法学习收敛速度慢的问题，而且提出了一种自适应调整网络参数的新算法，从而大大提高了算法的学习效率和综合性能．文中详细研究了基于改进的ＢＰ算法神经网络的实验仿真系统的建模过程，并在精细化工实验方案的分析和优选中获得了令人满意的效果．his paper not only proposes a new kind of activation functions, called segment activation function(SAF), but also presents an improved back propagation algorithm in which networks parameters can be adaptively adjusted. As a result, the learning efficiency and the comprehensive properties are greatly improved. Meanwhile, the modeling procedure for the experimental simulation systems which have been applied to the optimal selection for fine chemical experiment conditions is discussed in the paper. The application results have proved to be very satisfactory.

关 键 词：神经网络 活化函数 自适应 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]