Time series weighted prediction method using multikernel LS-SVR

Fault or health condition prediction of complex systems has attracted more and more attention in recent years. Due to complex dynamic behavior and uncertainty in running, complex systems are always difficult to establish precise physical model. Therefore, the time series of complex equipments are often used to perform the prediction in practice. In order to satisfy the requirements of application, which are good prediction accuracy and less calculation time, we utilize multiple relevant time series and propose a new prediction method based on multikernel LS-SVR. In this method, we proposed a simple computational method to obtain combining weights of multikernel, and the new prediction model considers the different effect of historical data into the prediction process. Prediction experiment is made by two relevant time series of one complex avionics equipment. The results indicate preliminarily that the proposed method is a practical and effective prediction method for its good prediction precision within less time cost.