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

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.