Advancing Drilling Safety: Automated Anomaly Detection in Well Control Using Machine Learning Techniques

Abstract The rise of anomalies like kicks, blowouts, lost circulation, and gas migration in drilling operations poses significant challenges to safety, environmental sustainability, and economic stability. Implementing frameworks for proactive monitoring and accurate anomaly detection is crucial to maintaining wellbore integrity, ensuring personnel safety, and minimizing environmental impact. This need is particularly acute in complex drilling environments, marked by intricate subsurface conditions and high costs, where unchecked anomalies can lead to severe consequences. Accordingly, this research emphasizes the importance of swiftly identifying and classifying such events, enabling timely interventions to prevent catastrophic outcomes and operational disruptions. This study introduces a multi-layered predictive model that effectively identifies and classifies well control anomalies, addressing the challenge of high false positive rates associated with existing research literature. This study utilizes a comprehensive dataset of historical well control incidents, including indicator parameters such as mud return rates, drilling fluid properties and wellbore pressure. The intelligent model is highly interpretable and outperforms existing counterparts in blind tests with a precision score of 0.918 and a low false positive rate of 2.38%, marking a significant advancement in intelligent anomaly prediction for drilling safety. This research improves traditional well control methods, which depend on equipment monitoring and slower responses, by employing real-time data analysis and machine learning. This shift provides drilling engineers with an advanced tool, enhancing safety and efficiency, and paving the way for more predictive and agile operations.