To improve the prediction accuracy of the International Roughness Index(IRI)of Jointed PlainConcrete Pavements(JPCP)and Continuously Reinforced Concrete Pavements(CRCP),a machine learning approach is developed in this...To improve the prediction accuracy of the International Roughness Index(IRI)of Jointed PlainConcrete Pavements(JPCP)and Continuously Reinforced Concrete Pavements(CRCP),a machine learning approach is developed in this study for the modelling,combining an improved Beetle Antennae Search(MBAS)algorithm and Random Forest(RF)model.The 10-fold cross-validation was applied to verify the reliability and accuracy of the model proposed in this study.The importance scores of all input variables on the IRI of JPCP and CRCP were analysed as well.The results by the comparative analysis showed the prediction accuracy of the IRI of the newly developed MBAS and RF hybrid machine learning model(RF-MBAS)in this study is higher,indicated by the RMSE and R values of 0.2732 and 0.9476 for the JPCP as well as the RMSE and R values of 0.1863 and 0.9182 for the CRCP.The accuracy of this obtained result far exceeds that of the IRI prediction model used in the traditional Mechanistic-Empirical Pavement Design Guide(MEPDG),indicating the great potential of this developed model.The importance analysis showed that the IRI of JPCP and CRCP was proportional to the corresponding input variables in this study,including the total joint faulting cumulated per KM(TFAULT),percent subgrade material passing the 0.075-mm Sieve(P_(200))and pavement surface area with flexible and rigid patching(all Severities)(PATCH)which scored higher.展开更多
基金supported by the Fundamental Research Funds for the Central Universities(Grant No.2021QN1006)Natural Science Foundation of Hunan(Grant No.2023JJ50418)Hunan Provincial Transportation Technology Project(Grant No.202109).
文摘To improve the prediction accuracy of the International Roughness Index(IRI)of Jointed PlainConcrete Pavements(JPCP)and Continuously Reinforced Concrete Pavements(CRCP),a machine learning approach is developed in this study for the modelling,combining an improved Beetle Antennae Search(MBAS)algorithm and Random Forest(RF)model.The 10-fold cross-validation was applied to verify the reliability and accuracy of the model proposed in this study.The importance scores of all input variables on the IRI of JPCP and CRCP were analysed as well.The results by the comparative analysis showed the prediction accuracy of the IRI of the newly developed MBAS and RF hybrid machine learning model(RF-MBAS)in this study is higher,indicated by the RMSE and R values of 0.2732 and 0.9476 for the JPCP as well as the RMSE and R values of 0.1863 and 0.9182 for the CRCP.The accuracy of this obtained result far exceeds that of the IRI prediction model used in the traditional Mechanistic-Empirical Pavement Design Guide(MEPDG),indicating the great potential of this developed model.The importance analysis showed that the IRI of JPCP and CRCP was proportional to the corresponding input variables in this study,including the total joint faulting cumulated per KM(TFAULT),percent subgrade material passing the 0.075-mm Sieve(P_(200))and pavement surface area with flexible and rigid patching(all Severities)(PATCH)which scored higher.
