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.
出处
《控制与决策》
EI
CSCD
北大核心
1996年第1期22-27,共6页
Control and Decision
基金
国家教委博士点基金
