Research On The Extreme Value Of The Vehicle Load Effect Based On The GPD Model

To ensure the bridge’s safety during its service period, it is necessary toconduct research on the extreme distribution of vehicle load effect. Somescholars have applied the reliability theory in this research field since1990, andup to know, many models have been proposed to describe the extreme vehicleload effect distribution treating the extreme value as a random variable. Withthe development of the detection technology and the installment of the WIM(Weight in Motion) instruments in more bridges, the information of local trafficflow becomes available. Using these data, we can investigate the characteristicsof the vehicle load effect of a bridge and predict further the extreme value overa longer period in future.Studies have shown that the distribution of the vehicle load effectdemonstrates multi-peak properties, and is difficult to be described by knownavailable probability models. After a thorough study on extreme value theory(EVT), a GPD (Generalized Pareto Distribution) model is proposed to fit thedistribution of daily maximum vehicle load. And later, the introduction of POT(Peak Over Threshold) theory enhances the power of GPD model to fit thedistribution of vehicle load effect.Investigations on the stochastic process of traffic load effect show that, thescheme to select load effect samples is error-prone, and it is necessary tostandardize the selecting procedure, otherwise, the selected samples may benon-iid (Independent and Identically Distributed). To solve this problem, theanalogous BM (Block Maximum) method is proposed in this study to guaranteethe iid property of selected samples, so that the GPD model could be used to fitthe distribution of the vehicle load effect directly.Besides, the estimator based on the space search is proposed by taking theKS statistic as an index of the fit-goodness, which ensures that the GPD modelemployed in this study is the most reasonable approximation of the observedsamples in the context of statistical significance. Meanwhile, the conception of search space is also significant to establish a framework for researchers toadopt results of others to achieve the optimized search space.To illustrate the application of the proposed GPD model, it is employed tostudy the probability distribution of vehicle load effect of a simply-supportedbeam using the recorded WIM data on site, and further predict the extremevalue over a longer period. Our results are compared with those acquiredemploying Chinese code, showing the reasonableness and weakness of thetraffic load model in Chinese code to determine the design load effect for theentire service life; moreover, for a shorter service life, the traffic load effecttends to be underestimated in most cases by Chinese code.