Ensemble and evolutionary prediction of layers temperature in conventional and lightweight cellular concrete subbase pavements

Extreme and fluctuating weather has a significant impact on the material properties of flexible pavements. Lightweight cellular concrete (LCC) can effectively mitigate weather effects due to its favourable insulating properties. To date, there has been little research on predicting temperature for different layers of conventional and LCC subbase pavements. This study investigates the application of LCC as a subbase material and its impact on layer temperature. Temperature profiles of two test roads, Erbsville and Notre Dame Drive (NDD), in Canada, have been collected for evaluation. Extreme gradient boosting (XGBoost) and genetic programming (GP) models were employed to forecast layer temperatures of Erbsville control and LCC-subbase sections based on inputs including ambient temperature, day of the year and constant depth. Shapley adaptive explanations (SHAP) were utilised for XGBoost, and parametric analysis was conducted for GP. Results indicated the superior performance of XGBoost (R-2> 0.98, MAE < 1.5 degrees C) over GP (R-2> 0.97, MAE < 1.87 degrees C), with both models demonstrating better predictive accuracy for LCC-subbase compared to the control section. SHAP, parametric analysis and external validation using NDD sections further validated the models' effectiveness in predicting temperatures for both control and LCC sections at various densities up to a depth of 0.8 m.