Probabilistic prediction method for seismic response and post-earthquake capacity of regional high-speed railway bridges

The rapid and accurate assessment of post-earthquake operational performance and the reliable planning of train operation schemes for a regional high-speed railway bridge network are fundamental to cross-regional disaster relief and rescue operations. This study proposes a method for constructing a regional bridge structure sample library based on Latin Hypercube Sampling (LHS) and Gibbs Sampling. A predictive model is developed to evaluate the seismic response of a regional high-speed railway bridge network, incorporating inputs from multiple feature sources. The seismic response limits and operational thresholds corresponding to various post-earthquake performance states of the bridges are quantified. A probabilistic assessment method for determining the seismic response of the regional bridge network is derived. In addition, an assessment method for post-earthquake capacity accounting for multiple sources of uncertainty, is established. The study demonstrates that the proposed sample library construction method for high‑speed railway structures requires at least 5 chains, each containing 8,000 samples. In the case study, the developed multimodal neural network surrogate model with embedded uncertainties successfully envelopes the true responses. Earthquake magnitude and epicentral distance are identified as the dominant factors affecting post‑earthquake operational performance, while topographic effects become significant when seismic intensity exceeds 7 degrees. Furthermore, the selection of decision criteria exerts a greater influence on optimizing post‑earthquake traffic routing than the choice of decision objectives.