一种改进的基于指数平滑神经网络模型的时间序列预测方法

An Improved Time Series Forecasting Method Based on Ewma-Ann Mixed Model

基于指数平滑模型与误差反传神经网络法提出了一个改进的时间序列预测方法.将神经网络模型移植入指数加权滑动平均模型中,充分考虑了时间序列的部分线性性和非线性性对预测结果的影响,是传统的混合模型的一个更合理的改进.最后通过对上证指数时间序列的实证分析,以预测均方误差为检验标准,对五种常用的时间序列预测模型进行了预测精度的比较,而且经验证所提出的改进的时间序列预测模型相对来说具有更小的预测均方误差.

The paper proposed an improved method of forecasting time series based on exponential smoothing and BP neutral network. The method had sufficiently considered influence of the linear and nonlinear part of time series on prediction result, which is a more reasonable improvement of traditional hybrid model. In the last part of the paper, we predicted time series of Shanghai composite closing index by using five models and compared the forecasting accuracy of each model by mean square error of each model as standards. Furthermore, a relative most mean square error had computed from the proposed time series forecasting method.