Seismic Risk Assessment of Regional Bridge Networks using a Stochastic Emulator Surrogate Model
研究使用高斯过程随机仿真框架建立单一替代模型，评估智利桥梁网络的地震风险，在保证精度的同时大幅降低计算负担，适用于区域级风险评估。
ABSTRACT In seismic-prone regions, assessment of the risk of regional bridge networks is essential for quantifying resilience metrics. Ideally this assessment needs to rely on high-fidelity numerical simulation models and on a comprehensive quantification of the seismic hazard. Unfortunately, the computational burden associated with such a faithful assessment of risk across the bridge portfolio is prohibitive for larger networks. This study investigates the use of a Gaussian Process (GP)-based stochastic emulation framework to alleviate this burden. A single surrogate model is established across all bridges of the same typology in the network. The surrogate model development accounts for diverse bridge configurations, as well as aleatoric ground motion variability across different intensity levels. The implementation also addresses aleatoric correlations across different engineering demand parameters for each bridge. The framework is implemented for risk assessment of the Chilean bridge network, specifically for four different bridge typologies, representing the predominant configurations within the national bridge inventory. It also examines in detail the computational benefits of establishing a single surrogate model across each typology, by comparing its accuracy to an application that considers the development of separate surrogate models per bridge type. The case study validates the accuracy of the stochastic emulation framework and illustrates the computational benefits for regional risk assessment. The latter is accomplished by comparing the results to different approximate formulations with respect to the assumptions for modeling the bridges within the network or the fidelity for considering and propagating the different type of uncertainties impacting the risk estimation.