Local online time series prediction based on the residual compensation with online SVR

For complicated nonlinear and nonstationary time series prediction, the precision will be decreased if a faster processing speed is reached in online SVR. A new time series prediction method is proposed. Local SVR is firstly adopted to make prediction of the time series. Then the forecast residual with the real data stream is calculated. Finally, the residual is estimated with online SVR algorithm to compensate the predicted value with the local SVR. Experimental results showed that the proposed method outperformed the Online SVR.