一种用于非线性逼近的新型模糊神经网络

A New Fuzzy Neural Network for Nonlinear Approaching

针对一般非线性映射的逼近问题 ,提出用分域逼近的通用算法来实现全局逼近 ,并据此构造了实现该算法的新型模糊神经网络。通过仿真 ,将新型模糊神经网络和常用的 BP和RBF两种神经网络进行比较。结果表明 ,该新型模糊神经网络的非线性逼近能力明显优于后两者 ,且权值具有明显的几何意义 ,设计难度相对较小 ,可用于解决复杂非线性函数的逼近问题。

With an aim at a general approaching problem of nonlinear mapping, a general sub field approaching algorithm is suggested to realize a global field approaching, on the basis of which a new fuzzy neural network is configured to carry out the suggested algorithm. The authors compare this new fuzzy neural network with BP neural network via the simulation and RBF neural network. The results indicate that the approaching ability of this new FNN is obviously superior to the latter two, and the weights have distinct geometric meaning and the design difficulty is relatively small. Accordingly, this new FNN can be used to approach any complicated nonlinear functions.