基于自适应噪声的鲁棒BP算法及前馈网的鲁棒性分析  

A Robust BP Algorithm Based on Self Adaptive Noise and the Robustness Analysis to Feed Forward Net

作　　者：石俊[1] 徐海银[1] 陈幼平[1] 周祖德[1] 

机构地区：[1]华中理工大学机械学院数控所

出　　处：《系统工程与电子技术》1998年第4期47-51,共5页Systems Engineering and Electronics

基　　金：国防预研基金

摘　　要：本文首先从几何和物理的角度分析了网络带噪声训练后的鲁棒性，再据此提出了一种自适应的训练方法，并通过仿真实验，证实了该方法的有效性；文中还从数学的角度定量分析了前馈网络对于噪声模式的鲁棒性，给出了设计鲁棒性网络的一般性原则。In this paper,the robustness of training net with noise is analysed from the point of geometry and physics firstly, then presents an algorithm of training with self adaptive noisy samples. The validity of the algorithm has been verified by emulated experiments; Furthermore, the robustness to noisy input vector for a feed forward net is developed quantitatively by mathematical method; Sequentially, a principle of designing a robust feed forward net is proposed. [WT5”HZ〗

关 键 词：鲁棒性 自适应噪声 前馈网 神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]