Data-driven nonlinear seismic response prediction method for building structures by synthesizing structural metadata and seismic characteristics

This study proposed a data-driven prediction method combining structural metadata and seismic characteristics to effectively predict the nonlinear seismic responses of structures. The featured method derives structural metadata such as natural period (T), strength ratio (SR), and ductility (μ) from a large number of nonlinear structural systems, and combines them with seismic characteristics like spectral acceleration to build large-scale predictive model learning data. The method using a convolutional neural network (CNN) integrates the correlation between structural properties and seismic information in the form of a conditional vector to predict nonlinear seismic responses and to simultaneously predict engineering demand parameters (EDPs), such as maximum inter-story drift ratio (MIDR) and peak floor acceleration (PFA). The performance of the prediction model was verified via nonlinear single-degree-of-freedom and multi-degree-of-freedom RC building structures. The developed model provided accurate and reliable predictions for various structures and seismic conditions. The proposed method is expected to be used as an effective tool for supporting decision-making when establishing seismic loss assessment and structural seismic retrofitting strategies.