Research On Prediction Method Of Expressway Pavement Performance In Sichuan Province Based On Machine Learning Combination Mode

In recent years,with the serious early damage of highway pavement and insufficient funds for road network maintenance,how to use the limited funds to effectively maintain the highway and realize the maintenance and appreciation of huge highway assets has become a major problem in the field of highway maintenance.Combined with the concept of big data,this paper collects and collates the pavement construction data,environmental climate data,traffic volume data,pavement performance test data and pavement maintenance history data of 23 expressways in Sichuan Province from 2014 to 2019,and establishes the pavement performance database.The first mock exam is conducted to study the factors affecting the pavement performance of highway.The effective factors of road performance are analyzed.The prediction model of pavement performance is built based on BP neural network,RBF neural network and support vector machine respectively.Principal component analysis,genetic algorithm and least square method are adopted to predict the XGBoost model.Through the improvement,the combined model has the advantages of fast convergence,high prediction accuracy and strong generalization ability.According to this,the pavement performance can be accurately predicted,the scientific pavement maintenance decision can be made,the maintenance funds can be reasonably allocated,and the economy of expressway maintenance can be improved.The main achievements are as follows:(1)In order to improve the quality of data,we collected and sorted the data of pavement construction,environment and climate,traffic volume,pavement performance test and pavement maintenance history of 23 expressways in Sichuan Province from 2014 to 2019,identified the abnormal data by box chart analysis method,and effectively interpolated the missing data by multiple interpolation method By calculating Pearson correlation coefficient,the correlation of pavement performance characteristic data is analyzed,which provides data support for the selection of subsequent algorithm model.On this basis,the sorted pavement performance related data information is imported into SQL Server management platform,and the highway pavement performance database of Sichuan Province is constructed.(2)Based on the idea of decision tree algorithm,random forest algorithm model and xgboost regression algorithm model are constructed to systematically study four kinds of pavement performance,including pavement damage condition,pavement driving quality,pavement skid resistance performance and pavement rutting condition.The main influencing factors of pavement performance are screened out,and the main influencing factors are sorted according to the difference of importance It is found that the calculation error of xgboost regression algorithm model is smaller,and the research conclusion of influencing factors of pavement performance is more reliable.(3)Combined with the road performance data of 23 expressways in Sichuan Province from 2014 to 2019,the road driving quality prediction model is constructed based on BP neural network algorithm and RBF neural network algorithm.In view of the relatively slow convergence rate of neural network,the principal component analysis algorithm is introduced to reduce the dimension of model input variables,simplify the model structure,avoid error superposition and improve the calculation efficiency At the same time,genetic algorithm is used to find the optimal model parameters based on the global,to avoid the occurrence of local optimal situation and improve the prediction accuracy of the model.After comparison and verification,it is found that the accuracy of the combined prediction models pca-ga-bp and pca-ga-rbf are 80.41% and 84.13%respectively,which can effectively predict the road driving quality.(4)Due to the common defects of neural network,such as slow convergence speed,model parameters easy to fall into local optimization and so on,the pavement performance prediction model is constructed based on support vector machine algorithm.In view of the complex and time-consuming calculation process of support vector machine,the algorithm model parameters are difficult to determine,the principal component analysis algorithm is introduced to reduce the dimension of the model input variables,eliminate the noise data,and simplify the calculation At the same time,genetic algorithm is used to determine the optimal model parameters to further improve the prediction accuracy.Through the prediction of the road quality of 23 expressways in Sichuan Province from 2014 to 2019,it is found that the pca-ga-lssvmr combination prediction model has the advantages of high accuracy It has the characteristics of fast convergence speed,less iterations and high prediction accuracy,and its prediction accuracy is 86%.It has good practical effect and can predict the road driving quality with high accuracy.