Data-Model Dual-Driven Intelligent Evaluation for Reservoir Development

During the process of oil reservoir development, especially in the middle and later stages of low to ultra-low permeability reservoirs, issues such as high water cut and production decline are particularly severe. To promptly grasp the current status of the reservoir and identify those with poor development outcomes, reservoir engineers need to regularly evaluate and grade the development effectiveness. This study integrates traditional petroleum engineering methods with machine learning techniques, particularly the Boosting ensemble learning method, to enhance the accuracy of reservoir development assessment. The research initially preprocesses missing values and outliers in the dataset, addressing the issue of imbalanced datasets caused by the scarcity of records with poor evaluations, by employing various resampling techniques. Subsequently, key reservoir development indicators were selected through feature engineering. Ultimately, five different Boosting ensemble classification algorithms—CatBoost, XGBoost, LightGBM, AdaBoost, and GBC—were utilized, and the model performance was evaluated using indicators such as AUC, MCC, and G-mean, with comparative analysis conducted on the application effects of these algorithms. The study’s findings indicate that when dealing with multi-class and imbalanced reservoir development datasets, the CatBoost and XGBoost algorithms outperform other algorithms. The classification model proposed in this research can effectively assist in dynamic reservoir analysis and provide decision support for predicting reservoir development outcomes, holding significant reference value for the application of artificial intelligence technology in the oil and gas sector.