Scholarlink

<p>In order to forecast spalling in the longitudinal joints of Continuously Reinforced Concrete Pavement (CRCP), this study presents a hybrid model that combines Support Vector Regression (SVR) and Particle Swarm Optimization (PSO). The model was trained and validated using data from the Long-Term Pavement Performance (LTPP) database, consisting of 33 sections and 395 observations. Through hyperparameter optimization using PSO, the SVR model achieved a mean RMSE of 17.41 and an R<sup>2</sup> of 0.90, outperforming traditional models such as Decision Trees, Linear Regression, and Artificial Neural Networks (ANN). A comprehensive sensitivity analysis was conducted to evaluate the impact of key hyperparameters (KernelScale, BoxConstraint, and Epsilon) on model performance, confirming that carefully tuned parameters significantly improve predictive accuracy. The model’s generalization capability was further validated through a train-test split evaluation, where the PSO-SVR model achieved an average RMSE of 22.63 and an R<sup>2</sup> of 0.80 on the test set, demonstrating strong real-world applicability. Feature importance analysis revealed that pavement age, annual average daily truck traffic (AADTT), initial International Roughness Index (IRI), and structural layer characteristics (L3 Type and Thickness) are the most critical factors influencing spalling deterioration. The proposed model offers high computational efficiency, with an average training time of 0.0129&#xa0;s and prediction time of 0.0009&#xa0;s, making it suitable for large-scale pavement management applications. The findings highlight the significance of data-driven approaches in infrastructure maintenance, providing valuable insights for proactive pavement management strategies.</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

Innovative hybrid model for spalling prediction in concrete pavement: SVR enhanced by particle swarm optimization

  • 期刊论文

  • 发表日期：

    2026年1月19日

查看全文
  • Ali Alnaqbi,
  • Ghazi G. Al-Khateeb,
  • Waleed Zeiada

摘要

In order to forecast spalling in the longitudinal joints of Continuously Reinforced Concrete Pavement (CRCP), this study presents a hybrid model that combines Support Vector Regression (SVR) and Particle Swarm Optimization (PSO). The model was trained and validated using data from the Long-Term Pavement Performance (LTPP) database, consisting of 33 sections and 395 observations. Through hyperparameter optimization using PSO, the SVR model achieved a mean RMSE of 17.41 and an R2 of 0.90, outperforming traditional models such as Decision Trees, Linear Regression, and Artificial Neural Networks (ANN). A comprehensive sensitivity analysis was conducted to evaluate the impact of key hyperparameters (KernelScale, BoxConstraint, and Epsilon) on model performance, confirming that carefully tuned parameters significantly improve predictive accuracy. The model’s generalization capability was further validated through a train-test split evaluation, where the PSO-SVR model achieved an average RMSE of 22.63 and an R2 of 0.80 on the test set, demonstrating strong real-world applicability. Feature importance analysis revealed that pavement age, annual average daily truck traffic (AADTT), initial International Roughness Index (IRI), and structural layer characteristics (L3 Type and Thickness) are the most critical factors influencing spalling deterioration. The proposed model offers high computational efficiency, with an average training time of 0.0129 s and prediction time of 0.0009 s, making it suitable for large-scale pavement management applications. The findings highlight the significance of data-driven approaches in infrastructure maintenance, providing valuable insights for proactive pavement management strategies.