A Hybrid ISSA-XGBoost Model for Predicting Wellbore Leakage

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence in Engineering
• Structural Health Monitoring

Background:

• Wellbores are critical underground structures susceptible to deterioration and safety hazards during drilling.
• Diverse formation conditions and drilling disturbances cause various wellbore leakage types, with fractured leakage being a key concern.
• Effective structural health perception and early warnings rely on multi-source sensor monitoring data.

Purpose of the Study:

• To propose and validate a novel wellbore leakage prediction method using an improved sparrow search algorithm (ISSA) optimized Gradient Boosting Decision Tree (XGBoost).
• To analyze the structural deterioration mechanism of fractured wellbore leakage.
• To enhance the intelligent health monitoring and early warning capabilities for underground wellbore structures.

Main Methods:

• Developed an improved sparrow search algorithm (ISSA) incorporating Sobol sequence initialization, opposition-based learning, and adaptive Levy flight.
• Integrated intelligent optimization techniques to refine position update strategies for discoverers, followers, and vigilantes within the ISSA.
• Applied the ISSA to optimize hyperparameters of the XGBoost model for wellbore leakage prediction, creating the ISSA-XGBoost model.

Main Results:

• The ISSA demonstrated significant advantages in optimization accuracy compared to classical machine learning algorithms.
• The ISSA-XGBoost model achieved an AUC improvement of 4.46%, with accuracy at 95.1%, precision at 94.9%, recall at 94.7%, and F1-score at 94.2%.
• The proposed model showed high accuracy and good generalization ability on different datasets for fractured wellbore leakage prediction.

Conclusions:

• The ISSA-XGBoost model provides a reliable and accurate method for predicting fractured wellbore leakage.
• The developed approach enables intelligent health monitoring and early warning systems for underground wellbore structures.
• This study offers a robust sensing data analysis scheme and technical support for hazard prevention in drilling engineering.