Prediction of Asphalt Pavement Rutting Depth Based on Multi-Model Fusion of Stacking Algorithm

Rutting is a serious issue in asphalt pavement, which may reduce the pavement driving quality and safety. Accurately predicting rutting depth is a crucial task in pavement engineering, providing crucial decision support for asphalt pavement design and maintenance. However, accurate prediction of pavement rutting still remains a significant challenge for pavement engineers. This research first selects the loading number, temperature, dynamic modulus, asphalt layer thickness, and base layer type and thickness as candidate features. Data preprocessing, including outlier handling and feature selection, is then performed. Finally, based on the stacking algorithm, a multi-model fusion approach for predicting rutting depth in asphalt pavements is proposed, using ridge regression (RidgeR), K-nearest neighbor (KNN), multilayer perceptron (MLP), and random forest (RF) models as base models, and support vector machine (SVM) as a meta-model. The model is optimized using a Bayesian optimization model. Results demonstrate the feasibility of using correlation analysis for feature selection. Seven features, including axle weight, upper layer temperature, and middle layer modulus, were selected as predictive features. While all the basic models achieved good prediction accuracy, the stacking ensemble model exhibited lower variance and bias, demonstrating superior generalization capability. The asphalt pavement rutting depth prediction method based on the stacking algorithm multi-model fusion proposed in this research can accurately predict the rutting depth.