基于模糊聚类的二型模糊神经网络系统辨识

二型模糊神经网络结合了二型模糊系统描述实际情况不确定性和神经网络的学习能力,在非线性系统的辨识中得到了广泛应用。二型模糊神经网络参数学习使用最多的是反向传播算法算法,该算法原理简单,易于实现。但是该算法对初值敏感,不合适的初始会导致算法收敛于非最优解或者发散。针对反向传播算法的这一缺点,提出了一种基于模糊C均值聚类的区间二型模糊神经网络辨识算法。该算法选择高斯型隶属度函数,将模糊C均值算法得到的聚类中心初始化高斯函数的中心,而高斯函数的宽度利用模糊C均值聚类算法的隶属度和中心求取。通过2个非线性系统的辨识效果表明,提出的辨识算法具有较高的辨识精度,收敛速度较快。

Type 2 fuzzy neural network identification based on fuzzy clustering

Interval type 2 fuzzy neural network was widely used in nonlinear identification, which combined the ability of handling uncertainly for type 2 fuzzy system and parameters learning in real application. The common adapted parameters learning algorithm is back propagation (BP), which is simple and easy to achieve. However, BP algorithm is sensitive to initial values, inappropriate initial results will lead to convergence to non-optimal solutions or divergence. In according with this default, we proposed an interval type 2 fuzzy neural network identification method based on fuzzy c means(FCM) clustering algorithm. The proposed method initializes center of gauss membership function by FCM and the width of gauss membership function will be obtained by membership degree and center got from FCM clustering algorithm. The identification results of 2 nonlinear systems show that the proposed algorithm has higher accuracy and faster convergence speed.