基于BP网络曲线拟合方法的研究

在利用BP神经网络进行曲线拟合时，为了解决如何确定BP神经网络隐含层神经元数问题，提出了一种新的快速构建BP神经网络结构的方法，即如何由输入层神经元数、输出层神经元数及样本点数来确定隐含层神经元数，同时针对在曲线拟合过程中经常出现的一些问题提出了解决方案。实验结果表明，该构建方法和改进方案在提高曲线的拟合精度、加快收敛速度方面收到了较好的效果．

Study on method of curve simulation based on BP network

When the BP network was used to simulate a curve, a problem of how to confirm the number of the hiding layer nerve cells for BP network occurs frequently. In order to solve the problem, a new method of how to construct rapidly the model of the BP network was brought forward. That is to say how to confirm the number of the hiding layer nerve cells for BP network according to the number of the input layer nerve cells, the number of the output layer nerve cells and the number of the point specimens. At the same time, some methods for improving the BP network training were presented. The results indicates that the method for BP network constructure and these methods are simple and convenient, which converges fast and improves the precision.