Research On Pavement Smoothness Prediction Based On Intelligent Algorithm

With the continuous development of road transportation in China and even globally,the mileage of road construction is gradually increasing,and the road maintenance part plays a vital role in the future.Among them,pavement smoothness is an essential index in road design,construction acceptance,and maintenance,which reflects the comfort of drivers and passengers and the safety and durability of the pavement.Vehicles driving on the rough pavement will produce additional vibration,accelerating the damage and destruction of vehicles and the road.At present,the International Roughness Index(IRI)is widely used globally as an indicator to measure road roughness.It has an important relationship with road service performance.The test section can be collected directly by using the roughness detector.Then,the collected data are integrated and analyzed to calculate the IRI value of the current section.For a long time,the roughness detector has been the mainstream measurement equipment of pavement roughness.However,the relatively high cost of the it is not conducive to the periodic and extensive measurement of pavement.To this end,researchers at home and abroad are also committed to developing methods that can predict IRI,mainly by collecting data on factors affecting IRI to establish an IRI mathematical model and predict IRI values according to the model to achieve indirect acquisition of IRI.Therefore,prediction accuracy is an important and challenging problem to judge whether the model can be applied in practice,which involves extensive data support and excellent algorithm optimization.Therefore,based on the summary of IRI prediction methods at home and abroad,this paper adopts the intelligent algorithm Ada Boost and random forest for machine learning and builds Ada Boost Regression,ABR,and Random Forest Regression(RFR)are two IRI prediction models.They rely on different data sets and intelligent algorithms.The ABR model was recorded using 4,265 historical observations from Long-Term Pavement Performance(LTPP),including asphalt pavement structure,climate,traffic,and pavement performance.Through the pre-processing of the original data and the training and configuration optimization of the model,the IRI prediction model is finally obtained.The results show that the R~2 and MSE of the ABR model on the test set are0.9571 and 0.0088,respectively.Therefore,the ABR model has high accuracy and predictability,and the overfitting of the model is well controlled.The influence degree of different variables on IRI and their correlation are analyzed so as to provide a reference for future road maintenance.In addition,an instant model,RFR,is also established.RFR model is different from ABR in that it is derived from vibration data generated during vehicle driving and processed by Fourier transform and Kalman data fusion to remove interference factors and expand input variables.In order to collect data more accurately,this paper designed a portable measuring device based on road response to collect vibration data and then built an IRI real-time prediction model based on a random forest algorithm.The results show that the R~2 and MSE of the RFR model on the test set are 0.967 and 0.036,respectively,which has high prediction accuracy.Finally,the two models are further studied and analyzed.Some shortcomings are pointed out,and prospects are made to provide some references for the follow-up research on related aspects.Figure[26]Table[10]Reference[80]...