基于机器学习的水性渗透结晶材料抗渗性能预测模型研究OA

Waterborne capillary crystalline materials can significantly enhance the impermeability of concrete by forming crystalline products to block capillary pores.To optimize their mix proportions and elucidate the roles of individual components,impermeability tests based on the Box-Behnken design were conducted to systematically investigate the effects of potassium aluminum sulfate dodecahydrate,sodium hydroxide,silane coupling agent,sodium silicate solution,and nano-silica.The results show that all components have significant effects on impermeability,exhibiting distinct synergistic and antagonistic interactions.At low levels of coupling agent or sodium silicate,moderate increases in aluminum salts or sodium hydroxide promotes gel densification.However,at high levels,performance gains saturated due to hindered ion migration.Based on the experimental data,four predictive models:particle swarm optimized Gaussian process regression(PSO-GPR),Gaussian process regression(GPR),support vector regression(SVR),and extreme gradient boosting(XGBoost)were developed and compared.The PSO-GPR model demonstrated superior predictive performance,achieving R2,EMA,and ERMS values of 0.9974,0.1084,and 0.1350 on the training set,and 0.9516,0.4024,and 0.5827 on the testing set,respectively.Cross-validation shows it achieves better prediction stability and overall performance.Response surface methodology(RSM)and SHAP analysis indicated that sodium silicate solution was the dominant factor,with consistent trends in secondary factor rankings.SHAP analysis also provided detailed insights into nonlinear effects and feature interactions.These findings demonstrate that the PSO-GPR model is an effective tool for predicting the impermeability of waterborne capillary crystalline materials and offers reliable guidance for performance enhancement and mix design optimization.