Surrogate modeling for identifying critical bridges in traffic networks under earthquake conditions

Abstract

Bridges are crucial for post-earthquake response. While seismic retrofit can improve bridge resilience, resource limitations necessitate prioritization of critical bridges for upgrades. Traditional methods for identifying critical bridges via seismic risk assessment involve computationally intensive traffic simulations. To expedite this process, this study proposes a “simulation-free” surrogate model using Markov random walk and random forest. Additionally, a combined bridge ranking method based on One-at-a-time, Sobol’ index, and Gini importance is introduced, benefiting from the rapidity of this surrogate model. Application to a case study of the San Francisco Bay Area Road network demonstrates a significant computational time reduction of 98% compared to simulation approaches and the ability to achieve good prediction performance with few training samples, reducing the effort in collecting training data and facilitating more rapid evaluation of the impact of bridges. Furthermore, the combined ranking method outperforms existing methods in identifying critical bridges for network performance enhancement.