摘要
针对BP算法存在的收敛速度慢等问题提出改进方案,修改其相关参数并且提出如何选择合适的隐藏层节点个数。同时针对学习样本数据的有限性、BP算法易陷入局部最小值和容易出现过拟合等问题进行了研究,提出了采用多重交叉验证的再改进BP算法。仿真结果表明,交叉验证BP算法提高了网络学习的效率。
BP algorithm exists against the slow convergence and other problem, in order to improve the program, the system amends its relevant parameters and presents the method to choose a suitable number of hidden layer nodes. Focus on learning ofthe limited nature of the data sample and BP algorithm easy to fall into the local minimum and prone to over-fit problems, the system presented by multiple cross-validation method. The simulation results show that cross-validation BP algorithm improved the efficiency of learning.
出处
《计算机工程与设计》
CSCD
北大核心
2008年第14期3738-3739,3742,共3页
Computer Engineering and Design
关键词
神经网络
BP算法
交叉验证
过拟合
隐藏层
误差函数
neural network
BP algorithm
cross-validation
over-fitting
hidden layer
error function
