摘要
分析了ＢＰ网络存在的主要问题及其产生原因，提出了改进算法ＢＰＧ，以共轭梯度方向代替梯度方向进行搜索，并在学习过程中采用不精确的一维搜索、限幅和条件轮回等措施．计算机仿真结果表明：改进的ＢＰＧ算法优于原ＢＰ算法．
Analyze the main problems of BP neural network and an advanced BPG algorithm is put forward which uses conjugate gradient direction to search,and uses some measures such as non-accurate linear search,limiting range and conditionly return.The computer simulation results declare that the advanced BPG algorithm is superior to BP algorithm.
出处
《暨南大学学报（自然科学与医学版）》
CAS
CSCD
1998年第1期118-123,共6页
Journal of Jinan University(Natural Science & Medicine Edition)
关键词
BP算法
共轭梯度
神经网络
误差反传算法
BP algorithm
conjugate gradient
convergence speed
local characteristics
