Research And System Implementation Of Friction Coefficient Prediction Of Asphalt Pavement Under Moisture Condition Based On IGWO-CatBoost

The skid resistance of asphalt pavement plays an important role in traffic safety.However,due to moisture condition in rainy days,skid resistance of asphalt pavement is significantly lower than usual,which induces a large number of vehicle skidding and rollover accidents,causing serious casualties and economic losses.Therefore,it is important to carry out research on the skid resistance of pavements under moisture conditions to reduce the traffic accident rate and road safety risks.In this thesis,the friction coefficient is used to evaluate the pavement skid resistance,and the prediction of friction coefficient under different moisture conditions is studied based on the texture morphology characteristics of asphalt pavement.The main research work of this thesis is as follows:(1)The texture data of asphalt mixture specimens are collected and the data quality is improved.Asphalt mixture specimens of different gradation are selected to obtain highresolution surface 3D(three-dimension)texture point cloud data by laser scanner.To solve the defect problem in the original 3D texture point cloud data,the least square method is used to suppress the tilt of point cloud firstly,and the average height fluctuation of point cloud is reduced by 24.98%.Secondly,the median filtering algorithm is used to filter the noise,and the average PSNR of the point cloud is increased by 88.13% after noise reduction.Finally,the RBFbased neighborhood interpolation algorithm is used to repair the missing value,its R-squared value has reached 0.902.(2)The texture feature data set is established through the characterization and feature optimization of macro and micro texture based on regional topography and section profile characteristics.In order to adequately describe the details of surface texture,macro and micro textures are separated based on the 3D texture point cloud data with improved quality,and 48 texture feature parameters are selected from two perspectives of regional morphology and section profile.Spearman coefficient correlation analysis and feature importance analysis based on RF-RFECV are used to sift out redundant and irrelevant features,and 21 texture features are selected eventually.(3)A prediction model of friction coefficient in moisture condition based on IGWOCatBoost is established.A digital pendulum instrument is used to collect the friction coefficient of specimens in moisture condition.Three traditional machine learning models,namely multiple linear regression model,SVR and BP neural network and two ensemble learning models,XGBoost and CatBoost are established.The experimental results show that the ensemble learning models have a higher prediction accuracy compared with traditional machine learning models,among which the CatBoost has the highest prediction accuracy.To solve the problem that the traditional grid search method is easy to fall into the local optimal solution,the initial part of the wolf class division of the gray wolf optimization is improved using Bernoulli mapping algorithm,and the iterative convergence part of the algorithm is improved using nonlinear convergence coefficients to achieve the improved gray wolf optimization algorithm IGWO,and the hyperparameter search optimization is performed on the CatBoost model to construct the IGWO-CatBoost model.The experimental results show that the IGWOCatBoost model can accurately predict the friction coefficient under moisture condition,and its R-squared value is improved by 11.29% and 9.13%,and MAPE value is reduced by 4.66% and3.04% respectively,compared with the XGBoost and GS-CatBoost models.(4)Design and implementation of IGWO-CatBoost based moisture condition friction coefficient prediction system.A total of four system modules are designed based on the analysis of the system requirements: 3D texture point cloud data quality enhancement module,macro and micro texture feature calculation module,IGWO-CatBoost based moisture condition friction coefficient prediction module and texture feature database management module,and a GUI desktop application for Windows platform is developed based on Py Qt 5.Finally,the reliability of the system is verified by system testing.The IGWO-CatBoost model overcomes the defects of poor fitting ability and insufficient generalization performance in the traditional methods.The research provides certain theoretical basis for non-contact evaluation of skid resistance of asphalt pavement under moisture condition,and have positive significance for realizing intelligent and efficient evaluation of skid resistance of asphalt pavement.