分段逆回归与神经网络组合建模方法

The Combined Modeling Method of Sliced Inverse Regression and Neural Networks

通过将人工神经网络技术和分段逆回归技术相结合,提出一种针对多因素非线性时间序列数据的预测建模方法。这种方法在保证原有函数形式不变的前提下,简化了输入变量,缩小了神经网络的规模,增强了模型的泛化能力。同时利用此方法建立北京市就业人口需求总量的预测模型,证实该方法的预测效果,并且为该方法运用于实际预测建模提供参考。

This paper integrated Artificial Neural Networks with Sliced Inverse Regression, and proposed a modeling (method) aiming to nonlinear multi-factor data indexed by time. This method can reduce the number of input variable while the function is still unchanged. So the scale of Neural Networks can be reduced and the generalization can be enhanced. At the same time, a numerical forecast model of social employment of Beijing was built based on this modeling method. The (forecast) result verified the validity and rationality of this method, and the case indicates how the method can be used in (practice.)