一个基于模糊神经网络的数据逼近和泛化建模方法  

AN APPROACH FOR MODELING OF APPROXIMATION AND GENERALIZATION BASED ON FUZZY NEURAL NETWORK

作　　者：乔志骏[1] 刘其真[1] 易维列[1] 何永保[1] 刘军[2] 林世雄[2] 

机构地区：[1]复旦大学计算机科学系,上海200433 [2]上海航天局第八设计部,上海200233

出　　处：《模式识别与人工智能》2001年第2期253-256,共4页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金

摘　　要：本文介绍一个基于模糊神经网络的数据逼近和泛化建模方法,定义了一种模糊系统动态调节神经网络的学习率,给出了用迭代自组织数据分析算法确定神经网络结构、初始化神经网络参数的方法。在雷达天线罩视线误差建模中的应用表明,这种方法加快了网络的收敛速度,避免了局部极值,具有较高的数据逼近和泛化能力。In this paper, a new approach for modeling of approximation and generalization based on fuzzy neural networks is presented. A fuzzy system is defined and used to adjust the learning rate of the fuzzy neural networks. The iterative self-organizing data analysis algorithm is applied to predetermine the rules and the parameters of membership functions of the fuzzy neural networks. The application of this approach to modeling radar radome aspect angle error show that, it can significantly simplify the design process, reduce training time and improve the performance of the network.

关 键 词：模糊神经网络 学习率 泛化建模 数据逼近 模糊推理 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]