基于牛顿插值和神经网络的时间序列预测研究  

Research of Time Series Forecasting Based on Newton Interpolation and Neural Network

作　　者：杜洋[1] 

机构地区：[1]辽宁科技大学电子与信息工程学院,辽宁鞍山114044

出　　处：《石油化工高等学校学报》2007年第3期20-23,共4页Journal of Petrochemical Universities

基　　金：国家自然科学资金项目(60474058)

摘　　要：在时间序列法基础上应用插值理论和神经网络建立一种新的预测模型。首先采用插值法拟合历史销售数据并求出大量的数据训练神经网络,弥补了历史数据缺乏的问题;然后用训练好的神经网络代替传统的最小二乘法拟合时间序列因素,从而求出预测值。仿真结果表明,此模型能够有效地改善模型的拟合能力并提高预测精度。A new prediction model was proposed by Newton interpolation and neural network based on time series. Firstly, history sales data were fitted by Newton interpolation and neural network was trained by mass data, which makes up for the deficiency of history data. Secondly, time series factors were fitted by substituting the trained neural network for traditional least--squares methods to obtain the predictor. Simulation result shows the fitting capacity and the prediction precision of the model is effectively improved.

关 键 词：牛顿插值 神经网络 时间序列 预测 

分 类 号：N945.24[自然科学总论—系统科学]