柔性神经网络及其应用

FLEXIBLE NEURAL NETWORK AND ITS APPLICATION

设计了带参数的单极性和双极性柔性Sigmoid函数的一种柔性前馈神经网络(FBPN),并给出了相应的学习算法.和普通的前馈神经网络(BPN)不同,FBPN不仅能学习连接权,且同时能学习柔性Sigmoid函数的参数,因此,它能根据学习样本集,为每一个隐含层和输出层单元产生合适的Sigmoid函数形态.一个算例和二个应用实例说明,柔性神经网络能提高BP网络的性能,并能较好解决不同领域中的分类与预测问题.

A flexible feedforward neural network (FBPN) using unipolar/bipolar sigmoid functions with a parameter and its learning algorithm is proposed in this paper. Being different from general feedforward neural networks (BPN), both the connection weights and the parameter of sigmoid functions can be learned in this FBPN. Thus, based on the learning sample set, the adaptive shapes of sigmoid functions in hidden-layers and in output-layer can be formed. Compared with BPN, the performance of this kind of network is improved. The results of a numerical example and two application cases show that this FBPN holds promising properties.