摘要
从极限角度证明了传统BP算法收敛速度慢、产生局部极小点的原因之一是该算法采用的均方误差准则并提出了一种基于熵函数准则的BP算法。
The mean square error(MSE) criterion of BP algorithm is proved to be slow in convergence and making the proceeding fall into local minimal. A new BP algorithm based on the entropy function criterion is proposed. Both the theoretical analysis and simulation demonstrate that the entropy function criterion is superior to the MSE cirterion.
出处
《北方交通大学学报》
CSCD
北大核心
1997年第5期543-547,共5页
Journal of Northern Jiaotong University
关键词
神经网络
BP算法
误差准则
熵函数
neural network\ BP algorithm\ entropy\ error criterion