Research On The Prediction Models Of The Continuous Reinforced Concrete Pavement Smoothness Based On LTPP

Continuous reinforced concrete pavement（CRCP）has the advantages of long service life,large bearing capacity and high comfort.It shows excellent performance under heavy load,overload and complex geological conditions.Hence,it is of great significance to ensure and improve the service performance of CRCP.Smoothness is an important evaluation index involving road running quality,safety and comfort in performance evaluation,as well as reflecting its overall disease condition.However,existing prediction models on the smoothness of CRCP have limitations such as low accuracy and complex operation.Based on this,this study proposed the prediction methods on the smoothness of CRCP based on hybrid machine learning models.1.Combined with the analysis of the influence mechanism on the smoothness of CRCP and the indexes for evaluating the smoothness of CRCP in mechanical-empirical pavement design guidelines,this study selected the initial International Roughness Index（IRI_I）,the percentage of transverse cracks（TC）,the number of perforations per kilometer（PUNCH）,the surface area of resilient and rigid repaired pavement（PATCH）,the age of pavement（AGE）,the freezing index（FI）and the percentage of subgrade material passing the 0.075 mm sieve(P₂₀₀)as the input indicators for evaluating the smoothness of CRCP,the International Roughness Index（IRI）as the output indicator and used statistical analysis,correlation analysis,factor analysis to verify the feasibility of selecting data and indicators.2.Proposed The hyperparameter optimization method based on the improved beetle antennae search（MBAS）algorithm.The traditional BAS search algorithm was improved by Levy flight and adaptive weight method.The results showed that the root mean square error（RMSE）values convergence efficiency of the models were faster,and MBAS was obviously effective for hyperparameter optimization of the models.3.Compare The prediction accuracy of the smoothness of CRCP by different models was.Firstly,the consistency between the predicted values and the actual values of four hybrid machine learning models was analyzed,and the results showed that the hybrid machine learning model of random forest and the MBAS（RF-MBAS）had the highest prediction accuracy,and the correlation coefficient（R）,the square of the correlation coefficient（R²）,the RMSE and the minimum iteration number of convergence（T）of the test set were 0.918,0.843,0.186 and 30,respectively.Then,the accuracy of CRCP smoothness prediction by RF-MBAS and mechanical-empirical pavement design guidelines was compared,and the results showed that,considering all data sets,the correlation coefficients corresponding to the fitting between the predicted values and the actual values of the two models are 0.958 and 0.733,respectively,which means that the prediction accuracy of CRCP flatness was significantly improved by RF-MBAS.4.Analyzed the sensitivity of all input variables and human intervention input variables to IRI of CRCP respectively.The results showed that P200 and PATCH were of high importance among the input variables that could be artificially interfered.Therefore,engineers should focus on theirs influence on the smoothness of CRCP.