A Hybrid ISSA-XGBoost Model for Predicting Wellbore Leakage.

As critical underground engineering structures, wellbores may suffer complex structural deterioration and hidden safety hazards may be encountered during drilling. Multi-source sensor monitoring data provides an effective data basis for structural health perception and early warnings for wellbore structures at risk. The inherent diversity of formation conditions and the dynamic disturbances during drilling jointly lead to the differentiated presentation of drilling loss types, among which fractured, permeable, and vuggy losses are the most typical. This paper focuses on fractured wellbore leakage, regards wellbore leakage as an important structural failure form of underground drilling engineering structures. In-depth analysis and research on the structural deterioration mechanism of wellbore leakage were conducted, and we propose a wellbore leakage prediction method based on the improved sparrow search algorithm (ISSA) optimized gradient boosting decision tree (XGBoost). First, the Sobol sequence is adopted to replace the random initialization strategy, combined with the opposition-based learning mechanism; then, an adaptive Levy flight search mechanism is introduced to dynamically adjust the population ratio of discoverers and vigilantes; finally, intelligent optimization technologies are integrated to reconstruct the position update strategies of discoverers, followers, and vigilantes, enhancing the optimization adaptability of the algorithm. Relying on multi-field sensor monitoring datasets collected from actual drilling engineering, this paper compares the proposed model with wellbore leakage prediction models built by classical machine learning algorithms, and verifies its generalization ability on different datasets. Experimental data indicate that the improved algorithm exhibits significant advantages in optimization accuracy, enabling the proposed model to achieve an AUC improvement of 4.46%, along with accuracy (95.1%), precision (94.9%), recall (94.7%), and F1-score (94.2%). On this basis, the ISSA was applied to the hyperparameter optimization of XGBoost, constructing the ISSA-XGBoost prediction model. The method has high accuracy and good generalization ability in fractured wellbore leakage prediction, and it can realize intelligent health monitoring of underground wellbore structures, including early warnings. This study provides a reliable sensing data analysis scheme and technical support for structural health monitoring and hazard prevention in drilling engineering.