成果归属作者：

李治平

成果归属机构：

能源学院

作者：

Cheng, Yishan ; Fu, Yingkun ; Li, Zhiping ; Hong, Kui

单位：

China Univ Geosci Beijing, Sch Energy Resources, Beijing 100083, Peoples R China;Beijing Key Lab Unconvent Nat Gas Geol Evaluat & D, Beijing 100083, Peoples R China;Univ Alberta, Dept Civil & Environm Engn, Edmonton, AB T6G 1H9, Canada;PetroChina, Beijing Digital Intelligence Res Inst Co Ltd, Beijing 102206, Peoples R China

关键词：

FLUID FLOWBACK; SHALE; IMBIBITION; FRAMEWORK; VOLUME; FATE; OIL

摘要：

Tight-gas wells are commonly reported with varying water-flowback volumes after hydraulic fracturing. The key features which determine the varying water-flowback volume remain unclear. Also, predicting water-flowback volume is of great importance for fracturing/production optimization and reservoir management in tight-gas fields. This study assembles a large dataset comprising 18 geological, fracturing, and flowback features from 3579 tight-gas wells, and presents a machine-learning (ML) workflow to predict water flowback volume for target wells in the Ordos Basin. The workflow mainly involves six steps: dataset assembly and preprocessing, correlation analysis, train/test split, algorithm testing, scenario analysis, and feature importance analysis. Application of the ML workflow demonstrated that the Random Forest algorithm outperformed Multiple Linear Regression, Neural Networks, Support Vector Machines, and XGBoost in both predictive accuracy and computation cost. Also, reliable predictions were achieved by using an optimal combination of key features, including total injected water volume, proppant amount, formation thickness, perforation depth, well type, liquid nitrogen volume, and permeability. Additionally, we employ the ML workflow to determine the optimized nitrogen volume for the fracturing treatments of target wells. This work provides a practical ML-based tool for predicting water flowback, and may help guide the optimizations of fracturing design and flowback strategies in the target field.

基金：

National Natural Science Foundation of China [52574055, U22B2073]

语种：

英文

DOI：

10.1016/j.geoen.2025.214190

来源：

GEOENERGY SCIENCE AND ENGINEERING,2026():.

出版日期：

2026-02

提交日期：

2025-10-19

引用参考：

Cheng, Yishan; Fu, Yingkun; Li, Zhiping; Hong, Kui. Leveraging machine learning to predict water flowback in tight-gas wells: Insights from low-pressure reservoirs in the Ordos Basin[J]. GEOENERGY SCIENCE AND ENGINEERING,2026():.