Study On Evaluation And Prediction Of Pavement Performance Based On Cloud Model And XGBoost

With the increase of traffic volume and pavement service life,the distress of asphalt pavement is deteriorated year by year.Scientific and reasonable evaluation and prediction of pavement service performance is an important basis for making pavement maintenance decisions.In this paper,considering the characteristics of ambiguity and uncertainty of pavement evaluation,a normal cloud model based on weighted fusion was developed for performance evaluation of asphalt pavement.Meanwhile,a predicted model of pavement performance based on XGBoost algorithm was built to explore the application of machine learning algorithm in pavement performance prediction.First of all,the development of the common national and provincial trunk highways in Guangdong Province was analyzed.Considering the influence of natural environment,climate and traffic volume,the representative asphalt pavements were selected as the research objects of pavement performance evaluation and prediction.Secondly,the influence of different weighting methods on pavement performance evaluation was analyzed.The result shows that the index weights of different weighting methods have great differences.The weight difference of the sub-index by information entropy is the largest,and the decay rate of the calculated PQI index is the fastest and pavement performance level is evaluated more rigorously.The improved entropy weight method considering special value and abrupt change of data reduces the difference of weights among the sub-indexes.The weighting method based on reward-penalty function is beneficial to enhance the role of indexes which weight and value are low in pavement evaluation.The weighting method based on subjective and objective combination is more scientific and reasonable.Thirdly,the method of weighted fusion based on cloud model was proposed by the phenomenon of cloud atomization.Based on the comprehensive weight by atomization process of cloud,the pavement performance was evaluated by the normal cloud model and grey clustering method.The evaluation results of the cloud model based on weight fusion better show the degradation trend to the next grade or even the next two grades than the evaluation results of the grey clustering method.The cloud model has more accurate and detailed grade evaluation ability,which can provide early warning for the potential pavement damage caused by the pavement degradation.Finally,the prediction models of PCI,RQI,RDI and SRI indexes were built based on XGBoost algorithm,and the reliability of the proposed model was verified by comparing with the random forest algorithm.The sensitivity analysis of influencing factors was conducted by SHAP.The results show that the performance of the XGBoost model is better than the random forest model.Specifically,the coefficient of determination(R2)of the XGBoost model is greater than 0.9,and its mean square error(MSE)is less than 1.7,indicating that the model has high accuracy and good versatility.The results of SHAP sensitivity analysis show that road age is the most important factor affecting road performance,followed by traffic volume.PCI and RQI indexes are relatively more sensitive to the influence of road age,while RDI and SRI indexes are relatively more sensitive to the influence of traffic volume and natural location of road.The road age and traffic volume are negatively correlated with pavement performance,while the surface thickness is generally positively correlated with pavement performance.