| 设计前馈神经网络结构的一种新方法 |
| |
| 引用本文： | 陆系群,余英林.设计前馈神经网络结构的一种新方法[J].华南理工大学学报(自然科学版),1998(1). |
| |
| 作者姓名： | 陆系群 余英林 |
| |
| 作者单位： | [1]华南理工大学无线电与自动控制研究所 [2]浙江大学计算机系 |
| |
| 摘 要： | 提出了一种设计前馈神经网络结构的新方法，该方法对网络隐层节点的输出作奇异值分解，根据奇异值大小的分布情况来决定隐层节点的个数，优化网络结构。这种方法同样也适用于有多个隐含层前馈神经网络的设计。应用这种方法来设计一个前馈神经网络并对一复杂信号建模和进行预测，能取得令人满意的结果。
|
| 关 键 词： | 神经网络 结构设计 奇异值分解 时间序列 预测 |
A NEW METHOD OF DESIGNING THE STRUCTURE OF FEEDFORWARD NEURAL NETWORK |
| |
| Lu Xiqun,Yu Yinglin.A NEW METHOD OF DESIGNING THE STRUCTURE OF FEEDFORWARD NEURAL NETWORK[J].Journal of South China University of Technology(Natural Science Edition),1998(1). |
| |
| Authors: | Lu Xiqun Yu Yinglin |
| |
| Abstract: | In this paper, a new method of designing a feedforward neural network is proposed. SVD is performed on a hidden layer's output matrix. Based on the distribution of the corresponding singular values, we can decide the number of hidden nodes of the corresponding layer, and optimize the structure of the feedforward neural network. This method is also adapted to multi_layer feedforward neural network. Moreover, this method is applied to design a feedforward neural network to model and predict a complicated signal. The results are satisfactory. |
| |
| Keywords: | neural networks structure design singular value decomposition time series prediction |
| 本文献已被 CNKI 维普 等数据库收录！ |
