摘要
提出了一种多层递归模糊神经网络(MRFNN),并提出混合混沌搜索方法用于网络学习。该网络融合了T-S模糊模型,在隶属函数层和规则层有局部反馈连接。网络的学习分为结构学习和参数学习两部分。结构学习确定隶属函数层和规则层的节点数;参数学习由混合混沌搜索方法完成,利用混沌搜索优化前件参数,同时利用最小二乘法实现后件系数更新。对非线性系统辨识进行,仿真实验并对连续搅拌釜式反应器系统建模,结果表明:本文方法能够有效捕捉系统的动态特性,所建模型具有良好的精度。
In this paper, a multiplayer recurrent fuzzy neural network （MRFNN） is presented and a hybrid chaotic search method is proposed for learning. MRFNN combines T-S fuzzy model with local feedbacks in the membership layer and the rule layer. The learning of MRFNN consists of the structure learning and the parameter learning. Structure learning defines the number of nodes in the membership layer and the rule layer; while parameter learning is performed by hybrid chaotic search method. The premise parameters are optimized by chaotic search, at the same time the consequent coefficients are updated by least square estimation. Simulations results of nonlinear function identification and CSTR system modeling show that the proposed methods can capture the dynamic characteristics of system effectively, and the built models have better precision .
出处
《华东理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第4期589-593,612,共6页
Journal of East China University of Science and Technology
关键词
递归神经网络
T-S模糊模型
混沌搜索
最小二乘法
建模
recurrent neural network
T-S fuzzy model
chaotic search
least square estimation
modeling
