Scholarlink

<p>Accurate prediction of rutting depth in asphalt pavements is critical for effective road maintenance and enhanced safety. To address the complexity and multifactorial nature of rutting phenomena, this paper proposes a hybrid prediction model that integrates mechanical-empirical approaches with advanced data-driven methods. Specifically, a multiscale mixed-kernel Gaussian process regression (MGPR) algorithm based on a combined kernel function optimized via a novel curvature-adaptive limited-memory BFGS with bound constraints algorithm is designed to enrich dataset quality. A revised mechanical-empirical model enhanced by three calibration parameters optimized through a multi-strategy adaptive particle swarm optimization (MAPSO) algorithm to ensure better convergence and global search capability. An elastic incremental broad learning system (EIBLS) utilizes elastic net regularization to address residual complexities, mitigate overfitting, and enhance generalization. Experimental validation using RIOHTrack (China’s first full-scale pavement test track) rutting data demonstrates that the integrated MGPR-EIBLS-MAPSO-RME model delivers superior prediction accuracy and robustness, offering significant potential for optimized pavement maintenance planning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid rutting depth prediction model integrating mechanical-empirical methods and data-driven approaches

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Zhuoxuan Li,
  • Jinde Cao,
  • Marek Kisiel-Dorohinicki,
  • Leszek Rutkowski,
  • Hanjie Liu,
  • Xinye Zhou,
  • Wei Huang

摘要

Accurate prediction of rutting depth in asphalt pavements is critical for effective road maintenance and enhanced safety. To address the complexity and multifactorial nature of rutting phenomena, this paper proposes a hybrid prediction model that integrates mechanical-empirical approaches with advanced data-driven methods. Specifically, a multiscale mixed-kernel Gaussian process regression (MGPR) algorithm based on a combined kernel function optimized via a novel curvature-adaptive limited-memory BFGS with bound constraints algorithm is designed to enrich dataset quality. A revised mechanical-empirical model enhanced by three calibration parameters optimized through a multi-strategy adaptive particle swarm optimization (MAPSO) algorithm to ensure better convergence and global search capability. An elastic incremental broad learning system (EIBLS) utilizes elastic net regularization to address residual complexities, mitigate overfitting, and enhance generalization. Experimental validation using RIOHTrack (China’s first full-scale pavement test track) rutting data demonstrates that the integrated MGPR-EIBLS-MAPSO-RME model delivers superior prediction accuracy and robustness, offering significant potential for optimized pavement maintenance planning.