摘要
BP算法是神经网络中最常用的算法之一。标准BP算法存在的最主要问题就是易于陷入局部极小、收敛速度慢等问题。针对BP算法的这些问题,出现了许多改进的措施,如引入变步长法、加动量项法等。提出了一种基于样本期望训练数的改进BP算法,仿真实验说明了该算法可以明显提高BP网络学习速度,并且具有简单通用性,可以和其他方法结合,进一步提高算法的收敛速度。
BP algorithm is one of the most widely used algorithms in neural network. In view of the main limitations of BP algorithm, such as easy to fall into local minimum value and slow in convergence, .several methods such as momentum term and variable step - size etc. are led to optimize BP algorithm. Proposed a new kind of improvement BP algorithm based on sample expected training number. Experiment results show that this algorithm has superiority in convergent velocity, simplicity and commonality.
出处
《计算机技术与发展》
2009年第5期103-106,共4页
Computer Technology and Development
基金
国家自然科学基金(60273043)
教育部博士点基金(200403057002)
安徽省自然科学基金(050420204)
关键词
神经网络
BP算法
样本期望训练数
收敛速度
neural network
BP algorithm
sample expected training number
convergent velocity
