Early detection of gas influx (or a kick) into the wellbore during drilling and completion operations is crucial for preventing uncontrolled release of hydrocarbon incidents that could lead to loss of lives, properties, and environmental contamination. In this study, we demonstrate the application of deep learning to automatically detect gas influx signatures across a 5163ft-deep wellbore. Optical fiber-based distributed acoustic sensor (DAS) data from eight well-scale tests were analyzed to investigate the accuracy of the automatic kick detection algorithm for gas influx volumes ranging from 2 to 15 barrels, wellbore circulation rates of 0 to 200 gallons per minute, and gas injection methods through the tubing or an injection line. The deep learning model uses convolutional autoencoders and effectively captures the gas signature for the eight datasets analyzed, providing an overall accuracy of 81% on the blind testing data (based on the structural similarity index measure). The results demonstrate an automated approach for gas kick detection to improve the safety of energy operations.
|