人工神经网络BP算法的改进和结构的自调整

The Improvement of BP Algorithm and Self-Adjustment of Structural Parameters

本文解决了ＢＰ神经网络结构参数和学习速率的选取问题，并对传统的ＢＰ算法进行了改进，提出了ＢＰ神经网络动态全参数自调整学习算法，又将其编制成计算机程序，使得隐层节点和学习速率的选取全部动态实现，减少了人为因素的干预，改善了学习速率和网络的适应能力．计算结果表明：ＢＰ神经网络动态全参数自调整算法较传统的方法优越．训练后的神经网络模型不仅能准确地拟合训练值，而且能较精确地预测未来趋势．

This paper resolves the problem of selecting structural parameters, learning rate and improves BP algorithm of artificial neural network, the self-adjusting algorithm of all parameters has been proposed for the back-propagation learning, and programmed a C language procedure. It can make the selection of hidden layer units and learning rate easily in the course of training, reduce external interference and improve the adaptive ability of learning rate and neural network. Our conclusion shows that the self-adjusting BP algorithm of all parameters is superior to the statistical modelings approach, the model of artificial neural network in tracing can not only exactly imitate training valuation but also make prediction accurately.