Toward prediction of petroleum reservoir fluids properties: A rigorous model for estimation of solution gas-oil ratio

Abstract

The amount of dissolved gas in production oil has been always a great question in oil and gas industry. Solution gas oil ratio is considered as a representative for the fraction of gas which is dissolved in oil during different stages of oil production. Several experimental methods have been developed for measuring this parameter. However, experimental procedures are usually time consuming, tedious and expensive. Thus, development of analytical equations and empirical correlations for estimation of solution gas oil ratio is of vital importance. In this study, a novel learning approach called least square support vector machine (LSSVM) optimized by coupled simulated annealing (CSA) was developed for calculating solution gas oil ratio as a function of temperature, bubble point pressure, gas specific gravity and oil API. To this end, a large number of data points including more than a thousand data sets from multiple reservoirs covering a wide range of reservoir conditions and pressure-volume-temperature (PVT) properties were gathered from various sources of literature. In addition, several statistical and graphical analyses including Average Absolute Percentage Relative Error (AAPRE), Average Percentage Relative Error (APRE), Root Mean Square Error (RMSE) and Coefficient of Determination (R2) were carried out to evaluate the accuracy and validity of model and to compare it with the most well-known implicit and explicit correlations of solution gas oil ratio estimation. Moreover, relevancy factor was employed to investigate the impact of each input parameter on solution gas oil ratio showing that bubble point pressure has the greatest effect on solution gas oil ratio. Finally, leverage approach was utilized to detect the data outliers and to find the applicability domain of the proposed model. The results in this study show that the developed model is able to estimate solution gas oil ratio with high accuracy and reliability making it possible to use the model in commercial software packages with various applications in oil and gas industry.