神经模糊系统中模糊规则的优选

提出一种基于两级聚类算法的自组织神经模糊系统，该系统采用两级聚类算法（改进的最近邻域聚类算法和Gustafson-Kessel模糊聚类算法）对输入／输出数据进行模糊聚类，并由模糊聚类的划分熵确定最优划分，建立模糊模型，模型精度可由梯度下降法进一步提高。仿真结果表明，这种神经模糊系统具有结构简单、规则数少、学习速度快以及建模精度高等特点。

Optimal choice of fuzzy rules in neuro-fuzzy systems

A self organizing neuro fuzzy system based on two stage clustering algorithm is proposed. Two stage clustering algorithm consisting of the nearest neighborhood clustering algorithm and Gustafson Kessel fuzzy clustering algorithm with cluster validity criteria is used to partition the input output space. The optimal number of fuzzy rules can be determined via fuzzy entropy as the criterion of cluster validation. A supervised scheme is utilized for constructing more optimal fuzzy model. Two simulation results show that the proposed method can provide optimal model structure and parameters for fuzzy modeling and possesses high learning efficiency.