Predicting and analyzing the cementing quality of oil well reservoirs based on Bayesian-random forest model

Abstract

Cementing quality crucially influences the safety, efficiency, and costs of oil and gas wells. Traditional evaluations rely on empirical and qualitative methods, introducing uncertainty. This study analyzed nearly 2000 wells casing and cementing data from 2016 to 2021 in an eastern China oil field. The original data was analyzed using multiple data cleaning techniques, and effectively dealing with missing values and anomalies. Additionally, the application of feature selection techniques such as random forest successfully identified key elements within the data. Three machine learning models and hyperparameter optimization methods were compared. A Bayesian random forest (BS-RF) model was established with hyperparameter optimization for cementing-quality evaluation. Finally, the relationship between the influencing factors and well-cementing quality were determined by the SHapley Additive exPlanations (SHAP) method. This enabled the evaluation and prediction of well cementing quality. The results indicate that the model achieved an accuracy of 93.21%. When the validation set was used for verification, the predictive accuracy was 95%. This indicates that the model is relatively stable and exhibits strong generalization capabilities across diverse datasets. Performance numeric of cement slurry (SPN), drilling fluid density before cementing, deviation angle, and flushing time are the four key features that influence cementing quality. Each has a SHAP value above 0.12. This article provides guidance for future well-cemented construction schemes.