结合SVD_TLS及EKF算法的动态自组织模糊神经网络在动态系统中的应用  

Dynamic Self-organizing Fuzzy Neural Network Combined with SVD_TLS and EKF Algorithm Used for Dynamic System Processing

作　　者：李云[1,2] 

机构地区：[1]中国矿业大学计算机科学与技术学院,徐州221116 [2]宿迁高等师范学院计算机学院,宿迁223800

出　　处：《计算机科学》2012年第B06期401-403,共3页Computer Science

摘　　要：针对如何优化模糊神经网络的规则及如何合理地调整非线性参数及线性参数等问题,提出了将奇异值分解_总体最小二乘法(SVD_TLS)及扩展卡尔曼滤波(EKF)相结合的动态自组织模糊神经网络(STD_DSFNN)。首先给出了STD_DSFNN的结构及各层的含义;其次,用EKF算法学习非线性参数,SVD_TLS算法学习线性参数的同时提取重要模糊规则;最后,通过典型的Machey-Glass时间序列预测实例验证SVD_TLS及EKF相结合的动态自组织模糊神经网络(STE_DSFNN),同时与DFNN、ANFIS及UKF_DFNN相对比,结果表明STE_DSFNN网络结构更紧凑,具有更好的泛化能力。This paper proposed SVD_TLS and EKF based dynamic self-organizing fuzzy neural network（STD_DSFNN）for optimizing fuzzy rules and Adjusting nonlinear and linear parameters reasonably.Firstly,the structure and meanings of each layer are given.Then nonlinear parameters are learned by using EKF algorithm,the linear parameters are learned by using SVD_TLS algorithm which also extract important rules at the same time.At last,the STE_DSFNN is verified through the typical Machey-Glass time series prediction examples.The results show that the STE_DSFNN network structure is more compact and has better generalization ability compared with the DFNN、ANFIS and UKF_DFNN.

关 键 词：奇异值分解_总体最小二乘法 扩展卡尔曼滤波 动态自组织模糊神经网络 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]