Physics-informed deep surrogate demand modeling for framed core-tube structures under seismic loadings
提出一种物理信息驱动的深度替代模型，快速预测高层框架核心筒结构在地震下的非线性响应，精度高且计算效率显著提升，适用于抗震性能评估和易损性分析。
Nonlinear time-history analysis of high-rise framed core-tube structures is computationally intensive, particularly when repeated simulations are required for seismic performance evaluation. This study develops a physics-informed deep surrogate framework for rapid prediction of multi-component nonlinear seismic demands. A parameterized family of structural models is generated within practical geometric and material ranges, and nonlinear bidirectional analyses under 500 ground motions are conducted to construct a comprehensive training dataset. Four engineering demand parameters—inter-story drift, top displacement, top acceleration, and base shear—are predicted in a sequence-to-sequence manner. The influence of temporal sequence length on predictive performance and convergence behavior is systematically examined. The optimal configuration achieves test-set coefficients of determination exceeding 0.93–0.95 across all response components. To enhance physical consistency, a hybrid loss formulation incorporating aggregate constraints on peak response, energy balance, and Arias intensity is introduced. While global statistical accuracy remains comparable to purely data-driven models, the physics-informed formulation reduces physically meaningful errors and improves training stability. Once trained, the surrogate enables millisecond-level inference, offering substantial computational savings relative to nonlinear finite-element simulations within the defined structural and seismic parameter domains, and provides a computational basis for future fragility- and reliability-oriented applications.