摘要
针对传统BP(back propagetion)算法存在的缺陷,分别对其收敛性标准、激活函数等进行改进,并采取措施防止振荡、加速收敛以及防止陷入局部极小.将改进后的BP神经网络运用到变形监测数据处理中,应用结果表明,改进后的BP神经网络比传统BP神经网络在精度等方面有了很大的改善.
This paper presents some improvements on the convergent criterion and activation function of the traditional BP neural network algorithm, and also the measures to prevent vibration, accelerate convergence and avoid falling into local minimum. The result of applying the improved network to deformation monitoring data processing indicates that the improved one is better than the traditional one in the aspect of precision.
出处
《同济大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第1期118-121,共4页
Journal of Tongji University:Natural Science
关键词
BP神经网络
遗传算法
激活函数
附加动量
变形监测数据处理
back propagetion neural network
genetic algorithm
activation function
attached momentum
deformation monitoring data processing
