Initial production prediction for horizontal wells in tight sandstone gas reservoirs based on data-driven methods

Area of Science:

• Petroleum Engineering
• Machine Learning Applications
• Reservoir Characterization

Background:

• Accurate initial production forecasting in horizontal wells is crucial for tight sandstone gas reservoirs.
• Traditional prediction methods are limited by reservoir heterogeneity and unfavorable petrophysical properties.
• Machine learning offers a promising alternative for enhancing prediction accuracy.

Purpose of the Study:

• To develop and validate machine learning models for predicting initial production in horizontal wells targeting tight sandstone gas reservoirs (IPHTSG).
• To systematically analyze engineering and production parameters for IPHTSG forecasting.
• To provide a data-intelligent decision-making framework for tight gas reservoir management.

Main Methods:

• Established an IPHTSG database by compiling engineering and production parameters.
• Reduced data dimensionality using correlation analysis and optimized model parameters via grid search and 10-fold cross-validation.
• Employed and compared six machine learning algorithms, with XGBoost selected for its superior performance.

Main Results:

• The XGBoost model achieved high accuracy (95% training, 93.33% testing) and strong performance metrics (precision, recall, F1-score).
• The developed models demonstrated reliability in predicting IPHTSG.
• The study identified key feature parameters including effective reservoir length, vertical thickness, and open-flow capacity.

Conclusions:

• Machine learning, particularly the XGBoost model, provides an effective and reliable method for IPHTSG prediction.
• This approach overcomes limitations of traditional methods in heterogeneous tight gas reservoirs.
• The methodology offers a replicable template for data-driven decision-making in optimizing development plans and production parameters.