Analysis of controlling factors for hydraulic fracturing parameters and accumulated production using machine learning

Abstract

This study, based on static data from over a thousand fracturing wells, employs data governance, data mining, and machine learning regression uncover principal controlling factors for production in the fracturing context. Preprocessing methods, including outlier identification, missing value imputation, and label encoding, address the field data challenges. Correlations among geological, engineering, and production parameters are analyzed using Pearson coefficient, grey correlation, and maximum mutual information. The AutoGluon framework and SHAP post-explanation method compute feature importance. Utilizing multiple evaluation methods, the entropy weight method comprehensively scores and ranks the principal controlling factors. A machine learning production prediction model is established for validation. Results show that DBSCAN achieves better accuracy in identifying field anomaly data.