Employing Novel Machine Learning to Achieve Precise Klinkenberg Slippage Factor Predictions in Gas Reservoir Characterization

Accurate prediction of the Klinkenberg slippage factor (KSF) is crucial for characterizing low permeability gas reservoirs, but conventional laboratory determination methods are prohibitively costly and time-consuming, constituting a major research gap. This study addresses this limitation by presenting a novel, experimentally validated workflow that seamlessly integrates comprehensive laboratory measurements with advanced machine learning (ML) models to achieve rapid and accurate KSF prediction. A robust dataset of 253 limestone core samples was utilized, employing air permeability, cation exchange capacity (Qv), porosity, tortuosity and grain density as input features. After a systematic comparative evaluation of ten algorithms on the capacity of ML to model the complex, non-linear relationships in KSF estimation are demonstrated. The key distinction and novelty reside in the superior performance of the AdaBoost algorithm, which achieved the highest predictive accuracy (R2=0.998, AARE%=1.276). This research offers a robust, efficient, and cost-effective alternative to traditional permeability correction techniques, establishing an innovative, data-driven framework that significantly enhances predictive modeling for reservoir evaluation in tight gas formations. • Predicting the Klinkenberg slippage factor in gas reservoirs • SHAP analysis to unveil the various effects of affecting parameters • Outlier detection via Leverage