摘要
本文介绍一种新的聚类方法 ,不需预先知道聚类数目 ,通过迭代运算使训练样本收敛到聚类中心 ,进而实现对样本的聚类 ,并给出了算法的理论证明 .将该算法应用到模糊神经网络中去 ,根据聚类结果建立一阶 TSK模糊神经网络 ,然后使用混合算法训练网络参数 ,分别用梯度下降法调整前提参数 ,递推最小二乘法调整结论参数 .最后 ,列举实例证明该算法的有效性 .
In this paper, a novel cluster algorithm is proposed to make training sample data converge to cluster centers via an iterative procedure without consideration of predeterminated center number. The novel cluster algorithm provides cluster centers and variances for Gaussian membership function to establish 1 order TSK fuzzy neural network. After establishment, a hybrid algorithm is implemented to tune network parameters, namely, to adjust premise parameters with gradient descent algorithm and consequent parameters with recurrent least squares respectively. Finally, simulation results are given to demonstrate the effectiveness of the implementation of this novel cluster algorithm in fuzzy neural networks.
出处
《信息与控制》
CSCD
北大核心
2002年第5期451-455,共5页
Information and Control
