| 正交投影神经网络的BP和GS杂交学习算法 |
| |
| 引用本文： | 肖少拥 石文俊 冯树椿 胡上序. 正交投影神经网络的BP和GS杂交学习算法[J]. 浙江大学学报(工学版), 2001, 35(2): 170-174 |
| |
| 作者姓名： | 肖少拥 石文俊 冯树椿 胡上序 |
| |
| 作者单位： | 肖少拥（浙江大学 计算中心, 浙江 杭州310027） 石文俊（浙江大学 计算中心, 浙江 杭州 310027） 冯树椿（浙江大学 计算中心, 浙江 杭州 310027） 胡上序（浙江大学 计算中心, 浙江 杭州 310027） |
| |
| 摘 要： | 主要讨论具有单隐层的正交投影神经网络的权值和阈值的学习问题，提出了一种新的将BP算法和GS算法相结合的杂交学习算法，其中GS算法对隐层到输出层的权值和阈值进行学习，BP算法用于输入层到隐层权值的学习，并给出一种最佳的隐层节点数的选取方法．仿真实验表明，该杂交学习算法具有学习速度快且能获得全局最优解的特点，并可有效地对学习过程中出现的病态情况进行求解，具有良好的普适性。
|
| 关 键 词： | 神经网络 杂交学习算法 BP算法 GS算法 病态问题 |
| 文章编号： | 1008-973X(2001)02-0170-05 |
| 修稿时间： | 1999-03-10 |
A hybrid algorithm basedon BP algorithm and GS algorithm of orthogonal projection neural network |
| |
| XIAO Shao|yong,SHI Wen|jun,FENG Shu|chun,HU Shang|xu,. A hybrid algorithm basedon BP algorithm and GS algorithm of orthogonal projection neural network[J]. Journal of Zhejiang University(Engineering Science), 2001, 35(2): 170-174 |
| |
| Authors: | XIAO Shao|yong SHI Wen|jun FENG Shu|chun HU Shang|xu |
| |
| Abstract: | Discusses the learning problems of weights and threshold of the orthogonal projection neural network with a single hidden layer, presents a new hybrid algorithm based on BP algorithm and GS orthogonal algorithm. The GS orthogonal algorithm is used to learn the weights and threshold of hidden layer and output layer, and the BP algorithm is used to learn the weights of input layer and hidden layer. The optimizing neuron's number of hidden layer for this hybrid algorithm is also given. The simulating experiments show that the hybrid learning algorithm can overcome the drawback of convergencing slowly and convergencing to loacl minima that BP algorithm has. It has the characteristics of learning fast and getting the global best solution, and it can obtain the solution when ill condition occurs, so it also has good generalization. |
| |
| Keywords: | neural network hybrid learning algorithm BP algorithm GS algorithm ill problem |
| 本文献已被 CNKI 维普 等数据库收录！ |
