Research On Pavement Performance Prediction And Premaintenance Timing Decision Method Based On Machine Learning

The study of highway pavement performance is a necessary tool to guide highway infrastructure construction and rational maintenance decisions,and the adoption of more reasonable maintenance decision tools can better adapt to the development of highway construction in the new era.Based on the pavement performance data collected from a highway in Shaanxi,this thesis systematically investigates the pavement performance decay prediction and maintenance decision problem during the whole life cycle with the goal of obtaining the best pavement premaintenance benefit and developing a machine learning-based pavement premaintenance optimal timing decision system in combination with the research results to achieve more reasonable supervision of the highway.The main research contents of this thesis are as follows:Firstly,to address the problem of constructing a highway asphalt pavement dataset,relying on the historical inspection data of a highway in Shaanxi,we analyze the spatial and temporal characteristics of the road condition level for indicators such as pavement damage condition and pavement flatness condition,and then improve the data quality through feature engineering methods such as outlier identification,outlier interpolation,feature construction,and feature coding,respectively,so as to establish a complete highway pavement performance dataset The data quality is then improved by feature engineering methods such as outlier identification,outlier interpolation,feature construction,and feature coding to establish a complete highway pavement performance dataset to provide data support for subsequent research.Secondly,for the highway pavement performance prediction problem,a PSO-XGBoost(Particle Swarm Optimization-e Xtreme Gradient Boosting)machine learning model based on particle swarm optimization is constructed based on the pavement performance detection dataset to obtain the global optimal particle position information to obtain the optimal hyperparameter combination of XGBoost model.The results show that the PSO-XGBoost model has the highest prediction accuracy with a coefficient of determination R~2 of 0.9336 and Mean Absolute Percentage Error(MAPE)of 0.3542%.Then,for the problem of choosing the best timing for preventive maintenance of the highway,the optimal timing decision method of pavement premaintenance based on the benefit-cost ratio was proposed.Based on the performance prediction results of the 15-year life cycle of this expressway,a decay model without and after maintenance is established,and the benefit-cost ratio method is used to calculate the final comprehensive benefit-cost ratio under different premaintenance timing in the whole life cycle by integrating the management department cost and user cost and to compare the maintenance benefit indexes under different premaintenance schemes for the best maintenance timing decision.Finally,based on the above pavement performance prediction and maintenance decision method,relying on the pavement performance testing data of a province in Shaanxi,the system functional modules were designed through user requirement analysis,and each functional module of the system was implemented based on Java framework and MySQL database technology,and the machine learning-based pavement performance prediction and premaintenance optimal timing decision system were developed,and finally,the software was tested from functional testing,performance Finally,the software system was tested from the functional test,performance test and compatibility test dimensions,and the results showed that the system met the design requirements and achieved the expected results.In this thesis,a complete pavement performance testing dataset is established with kilometer piles as the spatial granularity.Through data analysis and experimental validation,a PSO-XGBoost model is constructed to predict the pavement performance decay trend of the highway,and the results show that the proposed model has the highest prediction accuracy and makes scientific decisions on the timing of premaintenance of the highway based on the benefit-cost ratio.On this basis,an automated maintenance decision system is designed and developed,which can provide maintenance management departments with reasonable maintenance suggestions,thus saving maintenance funds and reducing maintenance resource consumption,and helping to realize scientific decisions on road premaintenance timing and providing a standardized management basis.