超球聚类模糊神经网络在跟踪控制中的应用

The Application of Fuzzy Neural Network Based on Super-Sphere Clustering in Tracking Control

采用简化的标准模型的模糊神经网络的基本形式而将模糊集的概念由一维推广到n维并利用超球聚类方法定义了一类n维模糊集及其隶属函数 ,提出一种基于超球聚类的模糊神经网络。这种模糊神经网络可以根据样本自动产生模糊规则 ,在一定程度上可避免“维数灾难”。将这种网络用于一类非线性系统的在线跟踪控制 ,定理表明 ,当聚类半径足够小时可使跟踪控制的静态误差任意小。

Taking the simplified COG (Center of Gravity) model of neural network as the elementary form, this paper presents a fuzzy neural network based on super-sphere clustering where fuzzy sets and their membership functions have been generalized from one - dimensional to n - dimensional. According to sample, this kind of fuzzy neural network can generate fuzzy rules automatically and can avoid 'dimensional disaster' to some extent. The new network is used in pursuit control of nonlinear system on line. The theorem shows that the static error of pursuit control can be small arbitrarily as long as clustering radius is small enough.