Abstract：The variation of the oil well production against time is controlled by several factors, in an extremely complicated nonlinear manner. The conventional petroleum reservoir engineering method suffers often from uncertainty of the correlation parameters and a large prediction error. Our analysis shows that the oil well production data have some features of time series. Therefore, the technique of the phase space reconstruction and the G-P method can used to obtain the optimal embed dimension and then to identify the time series of the well oil production. On this basis, by using the support vector machine method, the chaos-SVM model with time varying character is built,. with a very high precision for short-term well productions. In a real application, we can supplement the new historical data in real time to make the rolling prediction. The example W8-5 well application indicates that the relative error of the forecast results is only 7%, which shows that the coupling model has a good forecast ability and is of practical value.