Rapid seismic response prediction of city-scale RC frames under mainshock–aftershock sequences using deep learning and easily obtainable building information

Chenyu Zhang, Weiping Wen, Changhai Zhai, Guoqing Zhang, Nanqi Dai, Bochang Zhou,
Rapid seismic response prediction of city-scale RC frames under mainshock–aftershock sequences using deep learning and easily obtainable building information,
Structures,
Volume 82,
2025,
110777,
ISSN 2352-0124,
https://doi.org/10.1016/j.istruc.2025.110777.
(https://www.sciencedirect.com/science/article/pii/S2352012425025949)
Abstract: Mainshock–aftershock sequences can critically compromise reinforced concrete (RC) frame buildings, as initial mainshock damage is often intensified by subsequent aftershocks, undermining both structural safety and post-earthquake functionality. This study proposes a rapid deep learning–based prediction framework capable of estimating key seismic response indicators—peak inter-story drift ratios (IDR) and peak floor accelerations (PFA)—using city-scale easily obtainable building parameters (e.g., seismic fortification intensity, number of stories, story height) and mainshock–aftershock ground motion records. Compared with conventional nonlinear time-history analysis, the proposed method reduces computation time from minutes to milliseconds while preserving high prediction accuracy. Model interpretability is enhanced through ablation studies and SHAP-based feature importance analysis. Validation across multiple case studies demonstrates robust performance, achieving an average accuracy of 73.5 % for aftershock-induced damage. Beyond structural safety assessment, the method is further applied to post-earthquake damage evaluation and hospital resilience analysis, highlighting its capability to support rapid decision-making at the city-scale. The trained model, dataset, and graphical user interface (GUI) are publicly released, offering a practical and efficient tool for seismic risk assessment of building portfolios under mainshock–aftershock scenarios.
Keywords: Seismic response prediction; Reinforced concrete frames; Mainshock-aftershock sequences; Deep learning; City-scale resilience assessment