Tight sandstone reservoirs exhibit strong vertical heterogeneity and complex pore structures, challenging conventional permeability evaluation methods based on well logging data. While the rising machine learning (ML) techniques have demonstrated excellent accuracy for industrial applications, the physics and rationality within such a powerful “black box” remain less clear. Hence, reliable permeability prediction would benefit from an interpretable ML-based workflow that could reveal the controlling factors. To compare the models and examine the underlying features, 16 different ML sub-models are tested after data preprocessing, feature selection, and hyperparameter optimization. By comparing the fitting accuracy and tuning time, the light gradient boosting machine (LightGBM) optimized by the whale optimization algorithm (WOA), referred to as LGB-WOA, turned out to be the optimal model with the best fitting accuracy and relatively short tuning time; a field data application demonstrated that even in highly heterogeneous reservoir sections, the LGB-WOA model outperformed conventional petrophysical models by being the most consistent with reservoir permeability directly measured from core samples (R2 > 0.6). The SHAP (Shapley Additive exPlanations) values are then employed to interpret the predictions of our LGB-WOA model. As expected, the porosity curve (POR) exhibits the highest feature importance among all input features, significantly contributing to permeability predictions. Conversely, wellbore diameter (CAL) and compensated neutron log (CNL) contribute the least and need not be used for subsequent model improvements. The above experiments and workflow provide a powerful method for accurately assessing permeability in complex reservoirs and contribute to a broader understanding of the application of machine learning in reservoir characterization, paving the way for establishing more interpretable and reliable prediction models.
