Weighted prediction method with multiple time series using multi-kernel least squares support vector regression

Least squares support vector regression (LS-SVR) has been widely applied in time series prediction. Based on the case that one fault mode may be represented by multiple relevant time series, we utilize multiple time series to enrich the prediction information hiding in time series data, and use multi-kernel to fully map the information into high dimensional feature space, then a weighted time series prediction method with multi-kernel LS-SVR is proposed to attain better prediction performance in this paper. The main contributions of this method include three parts. Firstly, a simple approach is proposed to determine the combining weights of multiple basis kernels; Secondly, the internal correlative levels of multiple relevant time series are computed to present the different contributions of prediction results; Thirdly, we propose a new weight function to describe each data's different effect on the prediction accuracy. The experiment results indicate the effectiveness of the proposed method in both better prediction accuracy and less computation time. It maybe has more application value.