Resevoir simulations are routinely employed in the prediction of the performance of SAGD (Steam Assisted Gravity Drainage) process under different operating scenarios. They have been shown to have a significant potential to predict the future performance of SAGD operations. Due to the inherent uncertainty of petroleum reservoir data, the prediction needs to take into account the geological uncertainties associated with a particular reservoir. Usually, this is achieved by obtaining a history-matched model, conditioned to production field data. The model is then used to forecast future production profiles. Since the history match is time restricted to the previous production period, this is essentially an extrapolation problem with respect to time. Hence, such forecasts may not be very accurate. In addition, the simulation time is quite substantial for field studies with a large number of gridblocks. Therefore, forecasting workflows that reduce the number of necessary simulations while providing an accurate prediction are highly beneficial.
This paper presents a new approach for predicting production performance during SAGD process based on the results of SAGD simulations for a case with 3 well pairs. The approach utilizes the following workflow. Firstly, the field data were obtained, which consist of production outputs (particularly - cumulative Oil and cumulative SOR) for a typical time horizon of 5 years and the corresponding operating conditions. Secondly, a number of equiprobable geological realizations was generated using geostatistical methods to describe the permeability and porosity uncertainties of the reservoirs. Thirdly, direct numerical simulations of all realizations were conducted under a production operating scenario for a 10-year period. Then the data-driven proxy model is built that fits the actual field data to a linear, non-local function of the simulation data. The non-locality means that all the 10-year simulation results are considered in the match of the 5-year production data. After that the proxy is used to predict the production field data for the next five years. In this work, since the actual full 10 years of data are known, the predicted data are compared with the actual production data for the same period to evaluate the prediction quality.
This workflow is applied to a synthetic 3-Well-Pair SAGD model. It is shown that the proposed approach provides a highly reliable forecasting procedure for the reservoir considered. The difference between the predicted and actual field data lies within few percents, while the computational cost remains quite low. The use of the proposed approach in the prediction of uncertain reservoir performance under different operating scenarios during the SAGD process is also discussed.