摘要
针对标准的BP神经网络收敛速度慢、容易陷入局部极小值从而使网络误差增大的问题,提出了引入动量项的变步长BP网络预测算法。该算法在引入动量项因子的基础上,改变学习速率的步长,将改进的BP网络算法应用在预测上,通过仿真实验,得出了传统BP算法与改进BP算法的误差性能曲线。仿真验证表明,改进的BP神经网络有效地加快了收敛速度,而且使网络误差避免陷入局部最小值,预测效果更好。
In order to improve the speed of convergence and avoid the minimal value by standard BP network,a prediction algorithm of BP network,whose learning rate step was changed based on the introduction of momen-tum factor,was put forward.Getting the algorithm of the improved BP neural network applied in the forecast, the error performance curve of the traditional BP and the improved were obtained through the simulation experi-ment.The results showed that the improved BP neural network could effectively accelerate the convergence rate,and also made the network error avoid falling into the local minimum.
出处
《探测与控制学报》
CSCD
北大核心
2015年第5期102-105,共4页
Journal of Detection & Control
基金
山西省研究生优秀创新项目(20133104)
中北大学第十一届研究生科技基金(20141153
20141155)
关键词
BP
神经网络
动量项
步长
收敛
预测
BP neural network
momentum factor
step
convergence
forecast
