A mixture peaks over threshold approach for predicting extreme bridge traffic load effects

Abstract

Traditionally, bridge traffic load effects are considered as idenpendent and identically distributed random variables. However, load effects resulting from different loading events in terms of simultaneously involved vehicles/trucks do not have the same statistical distribution. To consider this, a novel method named mixture peaks-over-threshold approach is developed for predicting characteristic values and maximum value distributions of traffic load effects on bridges. The proposed method is based on the conventional peaks-over-threshold method, which uses the generalized Pareto distribution. The principle is to (1) separate the traffic load effects by types of loading event, (2) model the upper tail of the load effect for each type with generalized Pareto distribution, and (3) integrate them together according to their respective weights in the total population. Numerical studies have been conducted to demonstrate the feasibility of the proposed method in predicting characteristic value or quantile and extreme value distribution for bridge traffic load effects. Results show that the proposed approach is efficient to conduct extreme value analysis for data having mixture probability distribution function.