Data-driven model for predicting production periods in the SAGD process

Abstract

Many studies have analyzed the cumulative production performance in the SAGD (steam assisted gravity drainage) process by data-driven models but a study based on these models for a dynamic analysis of a SAGD production period is still rare. It is important for engineers to define the production period in a SAGD process as it has a stable and high oil production rate and engineers need to reset operational conditions after the production period starts. In this paper, a series of SAGD models were constructed with selected ranges of reservoir properties and operational conditions. Three SAGD production period parameters, including the start date, end date, and duration, are collected based on the simulated production performances. artificial neural network, extreme gradient boosting, light gradient boosting machine, and catboost were constructed to reveal the hidden relationships between twelve input parameters and three output parameters. The data-driven models were trained, tested, and evaluated. The results showed that compared with the other output parameters, the R 2 of the end date is the highest and it becomes higher with a larger training data set. The extreme gradient boosting algorithm is a better choice to predict the Start date while the artificial neural network generates better prediction for the other two output parameters. This study shows a significant potential in the use of data-driven models for the SAGD production dynamic analysis. The results also serve to support the utilization of the data-driven models as efficient tools for predicting a SAGD production period.