STATISTICAL MODEL ESTIMATION OF EXTREME THERMAL GRADIENTS IN LONG-SPAN BRIDGES COMBINING PARAMETER UPDATING

In order to establish the reliable statistical models for extreme thermal gradients in long-span bridges during their life-cycles, a generalized Pareto distribution (GPD) is proposed to describe the statistical features of thermal gradient samples that excess a threshold. And the procedure of excluding correlation in thermal gradient samples and the approach of selecting the best threshold are suggested. A Bayesian estimation method combining parameter updating for finding a GPD-based extreme thermal gradient model is developed to fuse prior information and incoming monitoring data. In this method, the Gibbs sampling is employed for computing the Bayesian posterior distribution. The developed method is verified by the thermal gradient data monitored on the Jiubao Bridge. The results indicate that the GDP has strong ability in describing the statistical characteristics of thermal gradients, especially in the tail region, and the Bayesian estimation method combining parameter updating provides more reasonable parameters for the GDP than the maximum likelihood estimation method. The outcomes of this paper are expected to offer a reference for the study of thermal gradients in long-span bridges.