摘要
本文针对传统BP算法在进行系统辨识时存在着速度慢、非平滑内插、受噪声影响很大、逼近精度不高等问题,并基于最大似然法对传统BP算法误差函数和能量函数及权值修正规则进行了修正,在此基础上提出了自适应鲁棒BP算法(ARBP),把它用于非线性系统动态辨识的应用研究中.仿真结果表明,ARBP算法能有效地进行非线性系统的动态辨识,并在逼近精度、收敛速度、防止量测噪声和过失误差对辨识影响等方面优于传统BP算法,从而为非线性系统的动态辨识提供了一个新的鲁棒算法.
The paper presents an adaptive robust BP algorithm(ARBP), which is aimed at the conventional BP algorithm that has some problems when it is applied to the dynamic identification, such as lower rate of covergence , nonsmoothing interpolation ,noise sensitivity and poor accuracy of identification. Based on the maximum likelihood method, the new algorithm has modified the error function and object function. It also updates the weights in the direction of steepest descent, but with a learning rate a specific function of the error and of the error gradient norm. The algorithm is applied to the dynamic identification of nonlinear system and the simulation rasult shows it is more efficient than the conventional BP algorithm on the aspects of the rate of convergence, accuracy of identification , resistance to gross errors and measuring errors. Thus, this is a new accelerated robust algorithm for the dynamic identification of nonlinear system.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1996年第1期78-86,共9页
Pattern Recognition and Artificial Intelligence
关键词
BP算法
非线性
系统辨识
ARBP
Adaptive,Robust BP Algorithm, Maximum Likelihood Method, Identification of Nonlinear System.
