摘要
以标准BP算法为基础,应用Levenberg Marquardt最优化方法,提出了一种快速收敛的BP算法———LMBP算法。经实验验证并与标准BP算法及其它改进形式比较,LMBP算法大大提高了收敛速度,而且性能稳定。这为BP神经网络应用于实时性要求高的场合(如在线检测)提供了算法基础。该算法的缺点是计算量大,所需计算机内存大,不适合大型网络的计算。
A new kind of BP algorithmLMBP that converges very fast is proposed in this paper by using LevenbergMarquardt optimization method and standard BP algorithm.The experimental results prove that LMBP converges very rapidly and has good stability property compared with that of the standard BP algorithm and other improved ones.LMBP algorithm is suitable for the case with high demands of online computation,e.g.online measurement.But when the size of neural network increases, it needs enormous calculation and large computer memory space.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2003年第4期79-84,共6页
Journal of Jilin University:Engineering and Technology Edition
