The hourly average solar wind velocity prediction based on support vector regression method

Abstract

A new neural network technique, support vector regression (SVR), is applied to forecast the solar wind (SW) velocity. SVR is a non-linear efficient tool for high data processing based on statistical learning theory. Its advantage is that the input only requires several periods data (about four 27-d solar-rotation periods to SW velocity prediction in this study), and the prediction is quite reliable. In our work, we deliberately choose the typical SW data covering all main space weather conditions: the SW data during the 9 yr from 1998 to 2006, which includes the periods of the SW speed variation associated with high-speed streams from coronal hole and coronal mass ejections. The performance of the SVR is measured by calculating the absolute average fractional deviation and correlation coefficient between the SVR model and observed SW velocity. We find that the predicted velocity values are over 90 per cent of the observed ones, i.e. the new approach is accurate and reliable in forecasting SW velocity. Based on the error difference, it can be concluded that the SVR technique can lend itself to future space weather forecasting models.