Hierarchical Bayesian modeling of highway bridge network extreme traffic loading

Abstract

The road network consists of bridges of various lengths and configurations, all of which require accurate prediction of traffic load within their lifetime. However, current prediction methods are limited to modeling and predicting traffic load for a handful of individual bridges only; no method can simultaneously model and predict the traffic load of all bridges within an entire road network. Further, conventional models neglect the information that exists in the traffic load effect data established for different bridges, leading to large estimation uncertainties for each bridge and load effect examined. This study proposes a hierarchical Bayesian model that can estimate the traffic load effect of multiple bridges simultaneously, and subsequently create predictions for the remaining (unexamined) bridges within the road network. The proposed model is demonstrated using the traffic load data and influence lines used in the background study for the Eurocode 1 Load Model 1. The results show significant reductions in prediction uncertainties, better fits as measured by leave-one-out statistics, more robust fits against extremes, and the emergence of intuitive correlation structures between different bridges’ traffic loads that are absent in conventional models. This paper also presents a potential new strategy to reduce estimation uncertainty, and a method to predict parameters and return levels for bridges across an entire network made possible by the proposed hierarchical Bayesian model.