Training Method of Support Vector Regression Based on Multi-dimensional Feature and Research on Forecast Model of Vibration Time Series

In recent years, Support Vector Regression (SVR) is used widely in predication field, with the advantages of structural risk minimization and strong generalization ability, which acquires good effects. The training characters of SVR model is the essential problem of affecting model accuracy. To solve the problem, this paper puts forward SVR model training method based on wavelet multi-resolution analysis, which adopts wavelet multi-resolution analysis to decompose time sequence and then uses the components data of each time spot as features to train SVR. The experiments has proved that the SVR training method which combines dynamic features of time series and detail information can improve the accuracy of the prediction model.