Statistical Analysis And Modeling For The Load Of Heavy Vehicle Queues

At present, many major bridge structures have installed the structural healthmonitoring system. As the growth of the monitoring time, data mining has attracted agreat deal of attention, due to the wide availability of huge amounts of data accumulatedby structural health monitoring system and the imminent need for turning such data intouseful information and knowledge to assess the safety of bridge structures.Given heavy trucks arrive in clusters have important influence on the overallreliability of bridge structures， this article focuses on statistical analysis andmodeling for the load of heavy trucks, especially heavy truck queues.First, this paper provides a simplified model for vehicle load analysis, which canconsider the effects of heavy vehicle queues. The gross vehicle weight set is dividedinto two parts, heavy vehicle set and light vehicle set, according to a certain threshold.If the gross weight of a vehicle is over the threshold, then it is belong to the heavyvehicle set, otherwise it is belong to the light vehicle set. The heavy vehicles areconsidered as masses, the light vehicles are weightless connections between twoadjacent masses. Cluster probability is defined to quantitative description the possibilityof two adjacent heavy vehicles should be considered as a queue. Both maximumlikelihood estimation method and Bayesian estimation method are developed for clusterprobability. In addition, analytical solution for cluster probability is derived by usingstochastic process theory.Second, dynamic linear model is used to modeling the time series of clusterprobability. Finite normal mixture models are used to fit the distributions of the totalgross weight of heavy vehicle queues and axis weight of trucks. Copulas are used tomodel the dependence structure of axis weight of trucks.Finally, different extreme value models are established for the total gross weightof heavy vehicle queues. A comparative study is carried on for extreme value predictionresults. For classical extreme value models often have no upper bound and can notprovide information of extreme value distribution，which will affect the accuracy ofvehicle load extreme value inference，a correction method is proposed in this article.