基于UKF的增长型模糊神经网络设计  

Growing fuzzy neural network design based on UKF algorithm

作　　者：韩红桂[1,2] 林征来 乔俊飞 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]计算智能与智能系统北京市重点实验室,北京100124

出　　处：《控制与决策》2017年第12期2169-2175,共7页Control and Decision

基　　金：国家自然科学基金项目(61533002;61622301);北京市自然科学基金项目(4172005);科技部水专项(2017ZX07104)

摘　　要：为了实现模糊神经网络结构和参数的同时调整,提出一种基于无迹卡尔曼滤波(UKF)的增长型模糊神经网络(UKF-GFNN).首先,利用UKF对模糊神经网络的参数进行调整;然后,设计一种基于隐含层神经元输出强度的模糊规则增长机制,实现模糊神经网络的结构增长;最后,将所提出的增长型模糊神经网络应用于非线性系统建模.实验结果显示,基于UKF的增长型模糊神经网络能够实现结构和参数的自校正,并且具有较高的建模精度.In order to adjust the structure and parameter of a fuzzy neural network simultaneously, a growing fuzzy neural network based on the unscented Kalman filter（UKF） method is proposed. Firstly, the UKF method is used to adjust the parameters of the fuzzy neural network. Then, a growing mechanism, using the output intensity of hidden neurons,is designed for self-organizing the fuzzy rules, and the structure of fuzzy neural networks can grow in the learning process. Finally, the proposed growing fuzzy neural network is used to model a nonlinear system. The experimental results show that the proposed growing fuzzy neural network is able to adjust the structure and parameters simultaneously,as well as with suitable modeling accuracy.

关 键 词：模糊神经网络 无迹卡尔曼滤波 增长型 非线性系统建模 

分 类 号：TP173[自动化与计算机技术—控制理论与控制工程]