Drilling Well Control Risk Assessment Method Based on Random Forest and its Application

As oil and gas exploration and development push into progressively deeper and ultra-deep formations, the overall risk profile of drilling operations rises markedly. Kicks represent one of the major hazards encountered while drilling oil and gas wells; consequently, rigorous well control risk assessment is the foundation for preventing, mitigating, and effectively managing kick incidents. Focusing on the characteristic features of drilling well control risk, this study systematically analysed kick case histories from the three perspectives of formation, wellbore, and wellhead. From this analysis, twelve well control risk indicators were extracted and organised into a comprehensive well control risk indicator system. Risk indicator rating and classification criteria were then established in accordance with industry standards for petroleum and natural gas well control. On this basis, introduced artificial intelligence techniques and employ an ensemble learning strategy, developed a random forest based well control risk assessment model. Model performance was evaluated using macro averaged metrics: the average precision (AP) and the average area under the receiver operating characteristic curve (AUC) across all risk categories. The model achieved an AP of 0.982 and an AUC of 0.989, indicating highly accurate predictive capability. Finally, the model was validated against data from three additional wells; the risk levels assessment results were in close agreement with field observations. The findings demonstrate that the proposed risk assessment model offers strong predictive power and practical applicability, providing field personnel with effective guidance for judging well control risk and informing timely operational decisions.