非线性系统神经网络辨识的鲁棒BP算法  

A Robust BP Learning Algorithm of Neural Networks for Nonlinear System Identification

作　　者：张邦礼[1] 李银国[1] 曹长修[1] 

机构地区：[1]重庆大学自动化系

出　　处：《控制与决策》1996年第1期22-27,共6页Control and Decision

基　　金：国家教委博士点基金

摘　　要：讨论系统辨识神经网络算法的鲁棒性问题。通过构造新的动态鲁棒目标函数得到的ＲＢＰ算法，能不断估计逼近精度，自动将品质好的样本置于强化学习域，并能有效地抵抗噪声干扰。实验结果表明，该算法具有鲁棒性强、收敛快、计算方便等特点。In this paper, the problem on robust learning algorithm of the neural networks for system identification is discussed. By constructing a new dynamic robust objective function, the RBP algorithm can continually estimate the approximation accuracy, and put the good samples into the domain of intensive learning, and can effectively resist to the noise perturbation. Experiment results show that the RBP algorithm is rubust,fast and convenient in iterative computation.

关 键 词：神经网络 系统辨识 鲁棒性 非线性系统 

分 类 号：O231[理学—运筹学与控制论]