Data-driven prediction of extreme response in high-speed railway bridges with bootstrap-based uncertainty analysis

This study explores the prediction of extreme structural responses in high-speed railway bridges using field-measured displacement data and statistical analysis. One-month data collected from instrumented bridges in Korea are analyzed using two methods: the Gumbel probability paper and Peaks Over Threshold (POT) approaches. Extreme mid-span displacements for 100-and 200-year return periods are estimated, and statistical uncertainty is evaluated via bootstrap resampling. To assess long-term performance, synthetic one-year datasets are generated based on the short-term records. Results show that the Gumbel method provides stable and consistent predictions, while the POT method is more sensitive to sample variability, particularly with limited data. However, both methods yield reliable estimates when sufficient data are available. This study offers practical insights into the application of extreme value theory for infrastructure monitoring and supports the development of data-driven strategies for resilient and sustainable bridge management.