Prediction of Mechanical Strength of AALRS under Extreme Lunar Environments: An Interpretable Stacking Ensemble Model Integrated with Hyperparameter Optimization Algorithms

Abstract

Abstract Alkali-activated lunar regolith is a promising extraterrestrial material for large-scale construction of lunar bases. However, the complex coupling effects of multiple influencing factors present significant challenges to strength prediction and mechanistic interpretation of alkali-activated lunar regolith simulant (AALRS) under extreme lunar environments. This study proposes an interpretable Stacking ensemble model integrated with hyperparameter optimization algorithms. The dataset includes features related to the lunar regolith simulant properties, mixture proportions, and environmental conditions that represent realistic lunar in situ construction scenarios. After tuning hyperparameters using six optimization algorithms, the Stacking model integrated with random search (RS) achieves the highest predictive accuracy and computational efficiency among the tested models [ R = 0.9482 , root-mean-square error ( RMSE ) = 5.58 MPa , mean absolute error ( MAE ) = 4.07 MPa , and computation time = 0.77 min ]. Feature contributions to strength and their interactions were analyzed using Shapley Additive Explanations (SHAP) and partial dependence plots (PDPs). Alkali content (AC), curing age (CA), and temperature were identified as the dominant features. Among all input features, the interaction between alkali content and initial temperature (IT) exhibits the highest interaction value. Parameters in the ranges of median particle diameter (MPD) < 100 μm , water-to-binder ratio (W/B) < 0.3 , CA > 7 days , AC > 0.06 , and IT > 60 ° C consistently exert positive contributions to predicted strength. Furthermore, a practical graphical user interface (GUI) based on the Stacking-RS model was developed to enhance user accessibility. This study establishes a data-driven paradigm that provides a generalizable framework for predicting and designing AALRS materials performance under extreme lunar environments.