摘要
针对BP算法这种当前前馈神经网络训练中应用最多的算法,在介绍BP神经网络的基础上,对标准的BP网络训练算法存在的收敛速度慢、易陷入局部极值的严重缺点,提出了几种学习算法上的改进;进而介绍了改进蹬算法在Matlab神经网络工具箱中的函数实现。最后应用实例利用Matlab神经网络工具箱对标准BP算法及改进的算法进行语言编程、仿真。仿真结果显示,改进后的算法在极值、收敛速度上得到了很大的改善。
BP algorithm is the most popular training algorithm for feed forward neural network learning. On account of the limitation of the standard BP algorithm, such as slow convergence and local minima, several improved algorithms are given and realized through the neural network toolbox in Matlab. Utilizing the neural network toolbox and programming, an example is given to show the differences between the standard BP algorithm and the improved algorithm using Matlab. The results show that the improved algorithm can improve in extremum and the speed of convergence.
出处
《控制工程》
CSCD
2005年第S1期100-102,共3页
Control Engineering of China
基金
浙江省科技计划资助项目(2004C31013)
关键词
神经网络
BP算法
改进算法
BP algorithm
neural network
improved algorithm
