基于模糊超球的模糊神经网络初始参数设定

A Enactment of The Primary Parameters Based on Fuzzy-exceeding Ball with Neuro-fuzzy Network

在人们常用到的模糊神经学习算法中很少讨论如何设置并调整初始的权值参数，这样就会影响模糊神经学习算法的准确度。使得目标函数值很大。虽然在改进的模糊神经学习算法中使用模糊C均值聚类方法来确定模糊规则的初始参数。但是这种方法必须已知模式集的数目，这就限制了模糊神经网络的应用范围。因此，将模糊超球神经网络的思想与模糊神经网络有机的结合起来，通过不断调整超球的中心和半径以及超球的数目，优化模式集的边界，来确定权值参数，利用这种方法确定初始参数可以减小误差，提高算法的准确度，使目标函数值减小。

Very few of the suitable initial values of tuning parameters are argued in neuro-fuzzy algorithm which are often used, so this can affect the mcety of neuro-fuzzy algorithm and increase the value of aiming function. Although we can design initial tuning parameters by using the fuzzy c-means clustering algorithm before learning the corresponding fuzzy rules, the number of pattern collection must be known firsdy.Thus it restricts the application scope of neuro-fuzzy network. Thereby, That will band the idea of fuzzy-exceeding ball with neuro-fuzzy network together, and adjust number ,centers and widths and of the ball, optimize the border of pattern collection to confirm the power values of parameters. We can minish error and improve nicety of algorithm by using it.