Hybrid approach for permeability prediction in porous media: combining FFT simulations with machine learning

Abstract

The prediction of permeability in porous media is a critical aspect in various scientific and engineering applications. This paper presents a machine learning (ML) model based on the XGBoost algorithm for predicting the permeability of porous media using microstructure characteristics. The seahorse optimization algorithm was employed to fine-tune the hyperparameters of the XGBoost algorithm, resulting in a model with predictive solid capabilities. Regression analysis and residual errors indicated that the model achieved good prediction results on the training and testing datasets, with RMSE values of 0.0494 and 0.0826, respectively. A SHAP value sensitivity analysis revealed that the essential inputs were the size of the inclusions, with the quantiles representing the maximum size of the inclusions being the most significant variables affecting permeability. The findings of this study have important implications for the design and optimization of porous media, and the XGBoost algorithm-based ML model provides a fast and accurate tool for predicting the permeability of porous media based on microstructure characteristics.