Study On Prediction Method Of Asphalt Pavement Service Performance Evolution Based On Multi-Source Sensing Data

At present,more than 50% of highways have entered the over-age service stage,and the service performance of pavement is evolving at an accelerated pace.The inability to accurately grasp the evolution process of pavement service performance will lead to the formulation of unreasonable maintenance plans,resulting in the waste of funds,inefficient disease treatment and increase of carbon emissions.It is of great research value for improving the scientific decision-making level of highway maintenance to excavate the evolution law of pavement service performance and build an intelligent hierarchical maintenance decision-making technology system.Driven by multi-source heterogeneous sensing data,represented by full-scale test ground and active highway data,systematic research is carried out on key technologies such as multi-source heterogeneous pavement detection and abnormal diagnosis and repair of pavement area environmental data,importance analysis of pavement performance impact factors,mining of pavement service performance evolution rules,and construction of hierarchical maintenance decision technology system,and a set of asphalt pavement evolution rules and maintenance decision methods based on artificial intelligence is proposed.The main research content is as follows:1.Construction of multi-source sensing data set for asphalt pavement service performanceFor the construction of multi-source sensing data set of asphalt pavement,the state information that has an impact on pavement service performance is collected from various pavement area environmental sensing data,mainly including pavement performance status,pavement structure information,pavement service life,traffic load,climate and other impact factors.At the same time,in order to solve the problem of multi-source heterogeneous sensing data fusion in pavement spatiotemporal domain,by mining the spatiotemporal distribution characteristics of various intelligent sensing data,aligning the spatiotemporal granularity,solving the problem of biased distribution and inconsistent accuracy level of spatiotemporal data across domains and systems,realizing the integration of multi-source heterogeneous intelligent sensing data of pavement,and completing the construction of multi-source sensing data set of asphalt pavement service performance.2.Design an anomaly detection and repair algorithm for multi type sensing data based on machine learningIn order to improve the quality of spatiotemporal sensing data of different types of asphalt pavement,three data anomaly detection and repair algorithms based on machine learning are designed.Compared with other detection and repair algorithms,the false detection and missing detection of abnormal data are reduced,and the repair accuracy is improved.The single dimension environment data anomaly detection and repair algorithm based on DS-LOF and Bi LSTM Attention achieves an average accuracy of 0.9590 for environment data repair;The multi-dimensional traffic load sensing data anomaly detection and repair algorithm based on SSC and GA-XGBoost achieves a repair accuracy of 0.9856 for traffic load data.The interpolation method based on spatiotemporal characteristics is used to repair road performance detection data.Through the analysis of the data stability before and after the repair by setting different anomaly ratio,when the anomaly ratio reaches 20%,the data repair can reduce the difference coefficients of the two types of sensing data by 6.07%and 8.32% respectively,which shows that the proposed repair method can improve the data quality,reduce data volatility,and provide reliable data support for the analysis of pavement intelligent sensing data.3.Propose the importance analysis method of impact factors of asphalt pavement service performance based on algorithm fusionAiming at the importance analysis of pavement service performance impact factors,the preliminary extracted pavement structural materials,traffic loads,meteorological and economic data are coded and normalized,and the many dimensional pavement service performance impact factors are analyzed using spearman correlation coefficient,recursive feature elimination(based on RF model)and XGBoost model.Then,fusion strategies based on the average importance weight and ranking voting are designed to fuse the results of the single model importance analysis to obtain more reasonable importance weight and importance ranking.Finally,based on the importance analysis results,the impact factors input into the subsequent prediction model are determined through manual review.4.Establishing a prediction model of asphalt pavement service performance evolution based on heterogeneous ensemble learningAiming at the prediction of the evolution of asphalt pavement service performance,the research is carried out based on the important impact factors of asphalt pavement service performance and historical detection data.By comparing the precision and error of grey model,neural network model,support vector regression model and multi class ensemble model in predicting the evolution of pavement service performance,the appropriate simple model is selected as the basic learner of the heterogeneous ensemble model.Based on the proposed ensemble strategy of precision forward weighting and error reverse weighting,a heterogeneous ensemble learning model of pavement service performance evolution law based on multi-dimensional and multi feature pavement sensing data is established.It is proposed that the average accuracy of the heterogeneous ensemble learning model is 0.005 and 0.011 higher than that of the simple average weighted model and the optimal simple model respectively,reaching 0.869.In terms of average error,the proposed model reduces the MAPE by 0.202% and 0.12% compared to the simple average weighted and optimal simple models,which can more accurately predict and analyze the evolution law of various pavement service performance indices,thus providing a data basis for intelligent hierarchical maintenance decision-making.5.Proposed an intelligent multi-level maintenance decision-making method for the whole life cycle based on cost benefit calculationAiming at the intelligent multi-level maintenance decision-making problem of asphalt pavement,based on the evolution law of various pavement service performance detection indices,taking the cost benefit ratio and preventive maintenance benefit into account,take PCI as examples to calculate the benefits of different maintenance decision-making methods.At the same time,according to the comprehensive evaluation index,damage condition,rutting depth,skid resistance and roughness index of asphalt pavement,an intelligent multilevel maintenance decision-making method covering daily maintenance,preventive maintenance,pavement surface function restoration,and pavement structural layer restoration is refined and proposed,including 7 types of maintenance decision-making and5 levels of maintenance schemes.The maintenance decision-making method is comprehensively optimized from multiple perspectives to maximize the cost-effectiveness of maintenance decision-making.In this study,100-meters pile is used as the minimum spatial particle size to excavate the evolution law of pavement service performance.In the heterogeneous learning fusion algorithm,parameters such as pavement structure and materials,traffic load,climate conditions,and pavement regular inspection indices are used to characterize the comprehensive pavement conditions.Through multi scene data and a large number of experiments,the research on the evolution law of pavement service performance and maintenance decision-making method based on multi-source sensing data proposed in this paper can refine the research content in the field of highway maintenance,to some extent solve the problems of low utilization of multi-source data,inaccurate prediction of pavement service performance evolution rules,untimely maintenance decisions,and difficulty in quantifying maintenance benefits in traditional maintenance decision-making,providing theoretical and technical support for highway maintenance decision-making.It also has important significance for improving maintenance efficiency and constructing long life pavement.