Research On The Prediction Method Of Frictional Characteristics Of Asphalt Pavement Surface Based On Gradient Boosting Model

The texture features of asphalt pavement are important factors that determine its skid resistance performance.During driving,the skid resistance performance of the pavement directly affects the traction and braking ability of the vehicle.Reasonable supervision of the texture features of the pavement can effectively improve its skid resistance performance and reduce the occurrence of traffic accidents.This paper addresses the problems of tedious traditional pavement texture feature acquisition,low accuracy of pavement skid resistance prediction models,and inadequate generalization ability.By using high-precision 3D laser scanning equipment to obtain the pavement’s 3D texture information,intelligent repair and feature extraction are carried out.A machine learning-based method for predicting the skid resistance performance of asphalt pavement is proposed.Firstly,this paper obtains the 3D texture data and friction coefficient of asphalt pavement by high-precision 3D laser scanning equipment and dynamic rotating friction coefficient tester,respectively.In order to identify and repair the outliers in the 3D texture data,KNN filling,Gaussian filtering and datum correction algorithms are used in turn to extract the macroscopic texture of the pavement using Fourier transform.The extracted texture features are fused with the speed and friction coefficient measured by the dynamic rotating friction coefficient tester,and then the Boruta algorithm is used to filter out the features with more important effects on the skid resistance performance,and to construct a prediction dataset of the skid resistance performance of asphalt pavement under different speed intervals.Next,three non-gradient boosting models,namely multiple linear regression,multilayer perceptron and random forest,which are widely used in the field of pavement skid resistance,as well as two gradient boosting models,XGBoost and Cat Boost,are constructed using the prediction dataset of asphalt pavement skid resistance in different speed intervals.The multimodel comparison experiments show that the gradient boosting model Cat Boost performs better than the non-gradient boosting model in the constructed asphalt pavement skid resistance prediction datasets at both low and high speed intervals,with R-squared of 0.9651 and 0.9643,respectively,so Cat Boost is selected as the final prediction model.Then,in order to further achieve the overall optimization of hyperparameters and improve the model prediction accuracy,this paper optimized the hyperparameters of the Cat Boost asphalt pavement skid resistance prediction model based on the Optuna hyperparameter optimization framework.Experiments show that the accuracy of the Cat Boost asphalt pavement skid resistance prediction model optimized by Optuna is further improved,with R-squared of0.9752 and 0.9800 on the asphalt pavement skid resistance prediction dataset at low and high speed speed intervals,respectively,which are improved by 0.0101 and 0.0157 compared with the results of grid search.Finally,based on the above research,a skid resistance prediction software integrating data quality improvement,macroscopic feature parameter extraction and multifunctional prediction of friction coefficient of asphalt pavement in different speed zones based on PYQT5 was designed,and the reliability test of the software was completed by black box testing.This thesis provides a theoretical basis for the prediction of pavement skid resistance based on asphalt pavement macrotexture and machine learning,and develops a software for predicting asphalt pavement skid resistance,which can have a positive impact on traffic road safety and has practical application value.