摘要
1 引言模糊系统建模一般将经过系统结构辨识和系统参数估计两个阶段。在辨识阶段,主要决定输入变量及其相互关系、模糊规则数、输入输出空间划分和系统参数的初值;在估计阶段,主要用来调整系统参数以使得系统的输出与目标输出的差值尽可能小。对于系统参数估计阶段的参数调整,人们已提出一些自动方法。对于系统结构辨识阶段,也产生了如模板法、聚类法和决策树法等,但这些方法一般都需要人工干预。其中模糊规则的生成与调整以及隶属度函数的选取是系统结构辨识阶段的主要问题,文提出了用神经网络自动生成模糊规则并进行隶属度形状调整,从而构成模糊神经网络。Wang提出自动分割输入空间的方法,Lin提出三阶段学习算法的模糊神经网络。
A new model of Fuzzy Inference Neural Network which employs Kohonen SOM to generalize the fuzzy rules is proposed in this paper. The network consists of three layers: the input-output layer, the If layer, and the Then layer. The performance of the Fuzzy Inference Neural Network is determined by the nodes and the connectivity between the layers. There are three learning processes in the network: the SOM process, the rule-generahze-and-abstract process and the LMS learnng processes. In the rule-generalization-and-integration process, another feature map is employed to abstract the rules effectively. The results of computer simulation applying to financial prediction show that the performance of this FINN network is superior to that of the BP network.
出处
《计算机科学》
CSCD
北大核心
2000年第5期61-63,共3页
Computer Science
基金
973国家重点基础研究发展规划项目
关键词
模糊神经网络模型
SOM规则
自动生成
Fuzzy Inference Neural Network, Self-Organization Map(SOM), Fuzzy rules, LMS learning
