A machine learning approach for estimating the bubblepoint pressure of world crude oils

Abstract

The pressure at which the first bubble of gas exits the reservoir oil is known as the bubblepoint pressure. This parameter affects multiphase flow in pipes and the overall recovery factor of oil from a reservoir. Therefore, it’s crucial to accurately estimate the crude oil bubblepoint pressure. There have been a lot of studies on calculating the bubblepoint pressure from laboratory data, which can be summarized into two main approaches: empirical correlations and machine learning (ML) algorithms. In this study, the authors implement both empirical correlations and ML algorithms with Decision Tree (DT), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and Group Method of Data Handling (GMDH). The data was collected from the open literature for world crude oils. The estimation results of the two approaches mentioned above are compared by regression metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Coefficient of Determination (R2). It was found that the GMDH algorithm has the accurate prediction results with the low MSE and RMSE (336605.4 and 580.177) and the highest R2 (0.9228). Trend analysis was carried out to strengthen model selection. The influence of input features on the prediction results indicates that the GMDH algorithm has the most stability. Therefore, the GMDH model is selected for estimating the bubblepoint pressure.