Application of Machine Learning as a Soft Sensor for Predicting Gas Wellhead Pressure: Case of Songo Songo Field

Machine learning has been increasingly used in the oil and gas industry to address various operational challenges, such as sensor malfunction leading to operational downtime. One approach to tackle this issue is by utilizing soft sensors, particularly for sensors that are difficult, expensive or important to obtain. In this study, a Machine Learning-based soft sensor was developed to predict the flowing wellhead pressure of the Songo Songo gas field, using a dataset of 16,747 rows of production data with 11 parameters. Two Machine Learning models, Extra Trees and Long Short-Term Memory, were trained using 80% of the dataset, then validated and tested with the remaining 20%. The Extra Trees model outperformed the Long Short-Term Memory model, with a mean absolute error of 0.01124, mean squared error of 0.00053, and root mean squared error of 0.02295. The developed Extra Trees model can predict wellhead pressure with an error within the recommended range of ±2% (ASME B40.100). The Machine Learning-based soft sensor with Extra Trees has demonstrated promising results in predicting wellhead pressure and can be integrated into the existing Songo Songo gas field Supervisory Computer-Aided Data Acquisition (SCADA) system to predict wellhead pressure that can alleviate downtime when a malfunction occurs.