组合ARMA与SVR模型的时间序列预测

Time Series Prediction Based on Mixture of ARMA and SVR Model

经典的ARMA模型常用于平稳时间序列的预测,而对于自然界绝大部分的非平稳序列一般采用确定性时序分析和随机时序分析。确定性时序分析对随机性信息浪费严重,而随机时序分析经过差分平稳序列后又回归到ARMA模型。本文利用在充分ARMA模型拟合后的残差序列进行支持向量回归(SVR)拟合,进而对原序列进行组合预测,比起单一模型的拟合及预测,该组合有效地提高了预测精度。

The prediction of stabled time series often use ARMA model, but to the most of the unstabled time series often use the analysis of certainty and randomness time series. It is a waste of the random information with the analysis of certainty, while after the difference of time series, it comes to ARMA model with the analysis of randonmess. The paper gives a way to full use of the residual data with SVR model after using ARMA model. Comparing to the single model used to predicting, the method of the mix- ture of ARMA and SVR model show to be more practical to improve the prediction precision.