Production-Increase Potential Evaluations after Refracturing Low-Shale-Oil-Producing Wells via Machine-Learning-Driven Multisource Data Mining

As shale reservoir development progresses, the share of low-producing horizontal wells grows, and the need for repeated-fracturing technology becomes more essential. Accurately evaluating the production potential is crucial for determining the efficacy of refracturing. This work proposes a novel method for assessing the repeated-fracturing potential of low-producing horizontal wells that combines the fine screening of important controlling parameters with an optimized XGBoost algorithm (<i>R</i> ² = 0.904 on test data). A multisource data set of 149 wells and 27 geological-engineering parameters is generated. Through a comparison of seven machine learning algorithms, the XGBoost algorithm outperformed the others in prediction performance. Six critical control parameters were found using feature priority ranking and variance inflation factor analysis: repeated-fracturing fluid injection volume and injection intensity, single-well-controlled geological reserves, fluid volume, length of the drilled oil-bearing formation, and number of repeated fracturing stages. Using the optimized XGBoost model, the potential of 29 additional candidate wells was assessed. The results indicate that two wells, X238-77 and Y3, exhibit considerable production increases and thus should be prioritized for repeat fracturing and reforming. Specific development plans for the remaining 27 wells are required according to their potential index rankings. This research provides a theoretical foundation and technical support for optimizing refracturing decisions, which is conducive to the efficient development of shale oil.