用于系统辨识的神经模糊推理系统的研究

通过具有代表性的、针对非线性系统辨识的例子，对自适应神经模糊推理系统在实际应用中采用不同方式的实现过程进行了详细分析与综合。首先，采用搜索树对模型输入变量的选取原则进行了分析；然后，分别采用聚类分割法与平均分割法对模糊模型结构辨识方面的各自特点进行了比较。在对辨识模型的验证中，重点分析了训练样本数目、模糊模型参数数目与模糊模型泛化能力之间的关系，并通过不同模型测试性能结果的对比，说明并联模型泛化能力的复杂性。

Study on Neural-fuzzy Inference System Used in the System Identification

This paper analyses in detail and synthesizes the different realization methods of adaptive neural-fuzzy inference system by means of the typical example of nonlinear system identification. First, the selection principle for the fuzzy wodel's number of input variables is analyzed by making use of the searching tree. Then, different features in the aspects of structure identification of fuzzy model using the clustering and means-clustering are compared. The relationships among the numbers of samples, fuzzy model parameters and the ability of generalization of the fuzzy model are analyzed, and the comparison of the test results of models shows the complexity of generalization ability of parallel model.