摘要
神经网络的非线性逼近能力的研究是神经网络成为辨识模型的理论基础。首先研究了基于正交多项式函数的神经网络逼近理论和方法,并在此基础上证明了新型Chebyshev神经网络具有良好的非线性并研究了它的全局最优逼近性质。然后提出了一种用于复杂非线性系统辨识的基于Chebyshev基函数的模糊神经网络模型和学习算法。该模型以Chebyshev基函数为隶属函数,规则后件采用输入变量的线性函数,无需调整隶属函数的参数,只是采用BP学习算法学习后件参数,因而大大减少了模型算法的计算量,学习算法简单,加快了学习收敛速度,而且不使网络结构复杂,设计简单。仿真结果表明所提模型和方法的有效性。
The theory of identification model based on neural networks（NN） is to research into its capability of nonlinear approximation. Universal approximation capability of orthogonal polynomials based on NN was proposed, and with which nonlinear approximation and global optimization of this new type Chebyshev NN were proved. Then fuzzy neural networks model and learning algorithm based on Chebyshev basis functions to be used as its membership functions were proposed for nonlinear system identification. As no parameters to be adjusted in the Chebyshev membership functions and just adopting BP algorithm studying parameters of fuzzy rules, the computing greatly reduced and the simple model structure, the fast convergence and the high precision of identification were obtained. The simulation results show their effectiveness of the proposed model and method.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2006年第3期590-593,共4页
Journal of System Simulation
基金
安徽省高等学校杰出青年教师资助课题(2005jq1119)
