Scholarlink

<p>Accurate surface roughness prediction is critical for effectively evaluating and planning maintenance for continuously reinforced concrete pavements (CRCP). This paper suggests a hybrid data-driven architecture that combines a Genetic Algorithm (GA) and a Gradient Boosting Machine (GBM) to estimate the International Roughness Index (IRI) of CRCP sections. 395 observations from 33 untreated CRCP sections were taken from the Long-Term Pavement Performance (LTPP) database. The modeling framework includes 19 explanatory variables for structural characteristics, conditions in the environment, and traffic loading, as well as two performance-related parameters: starting IRI and measured IRI. The GA was employed to optimize key GBM hyperparameters, including the number of learning cycles, learning rate, and maximum number of splits, within a five-fold cross-validation scheme. The resulting GA-GBM model achieved a mean Root Mean Square Error (RMSE) of 0.0369 and a coefficient of determination (R<sup>2</sup>) of 0.9922, demonstrating substantial improvement over benchmark models, including non-optimized GBM, linear regression, random forest, support vector regression, and artificial neural networks. Model interpretability was enhanced through feature importance analysis, normalized sensitivity analysis, and one-dimensional partial dependence plots, which consistently identified initial IRI, pavement age, L4 thickness, L3 thickness, annual average daily truck traffic, precipitation, and construction number as the most influential predictors of roughness progression. Overall, the proposed GA-GBM framework provides a robust and interpretable tool for high-accuracy IRI prediction and offers practical value for data-driven decision-making in CRCP performance management.</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

A Hybrid GA-Optimized Gradient Boosting Model for Predicting IRI in Continuously Reinforced Concrete Pavement

  • 期刊论文

  • 发表日期：

    2026年3月16日

查看全文
  • Ali Alnaqbi,
  • Ghazi Al-Khateeb,
  • Waleed Zeiada,
  • Khaled Hamad

摘要

Accurate surface roughness prediction is critical for effectively evaluating and planning maintenance for continuously reinforced concrete pavements (CRCP). This paper suggests a hybrid data-driven architecture that combines a Genetic Algorithm (GA) and a Gradient Boosting Machine (GBM) to estimate the International Roughness Index (IRI) of CRCP sections. 395 observations from 33 untreated CRCP sections were taken from the Long-Term Pavement Performance (LTPP) database. The modeling framework includes 19 explanatory variables for structural characteristics, conditions in the environment, and traffic loading, as well as two performance-related parameters: starting IRI and measured IRI. The GA was employed to optimize key GBM hyperparameters, including the number of learning cycles, learning rate, and maximum number of splits, within a five-fold cross-validation scheme. The resulting GA-GBM model achieved a mean Root Mean Square Error (RMSE) of 0.0369 and a coefficient of determination (R2) of 0.9922, demonstrating substantial improvement over benchmark models, including non-optimized GBM, linear regression, random forest, support vector regression, and artificial neural networks. Model interpretability was enhanced through feature importance analysis, normalized sensitivity analysis, and one-dimensional partial dependence plots, which consistently identified initial IRI, pavement age, L4 thickness, L3 thickness, annual average daily truck traffic, precipitation, and construction number as the most influential predictors of roughness progression. Overall, the proposed GA-GBM framework provides a robust and interpretable tool for high-accuracy IRI prediction and offers practical value for data-driven decision-making in CRCP performance management.