Real-Time Data-Driven Gas/Oil Ratio Estimation

Summary With this study, we propose a methodology to estimate in real time the gas/oil ratio (GOR) of oil wells using regression models based on pressure and temperature sensor data. We use kernel ridge regression (KRR) and support vector regression (SVR), leveraging their ability to capture nonlinear relationships between input variables. A real data set from wells in a pre-salt reservoir in the Santos Basin validates the methodology, showing that production test data can effectively train the models to predict GOR from pressure and temperature measurements. On average, the proposed models achieved a symmetric mean absolute percentage error (SMAPE) of 2.7% across all analyzed wells, demonstrating their accuracy. When applying an expanding window approach, the SMAPE was further reduced to 1.1%, reinforcing the models’ precision and robustness in estimating GOR. This methodology enables more frequent and precise GOR monitoring, enhancing oilfield operational efficiency and supporting data-driven decision-making processes. Its application allows for better reservoir management by anticipating significant variations in gas production and optimizing platform operations.