摘要
针对BP神经网络中学习因子取值小、收敛性好但训练时间长,学习因子取值大、权值变化剧烈但可能导致振荡的情况,提出了一种修正学习因子的方法,即给学习因子前加一比例因子,在网络权值调整过程中自动调整学习因子的大小,使网络训练时间短,而且收敛效果较好。仿真结果表明,在导弹指令跟踪中,改进算法比原来算法优越得多。
Seeing on that in BPNN the small learning gene will make the long training time, but the large learning gene will make the BPNN surging,this paper brings forward a way to modify the learning gene, that is, adding a proportion gene before the learning gene, The proportion gene will change when the weight of the BPNN needs to be modified. This can shorten the training time and make convergence better as well. The simulating results show that the new algorithm is much better than the old one during BPNN scouting the missile command.
出处
《中国工程科学》
2005年第5期63-65,共3页
Strategic Study of CAE
关键词
神经网络
改进算法
仿真
BPNN
improved algorithm
simulation