Structural seismic response prediction with LSTM-based Bidirectional Urban Safety Network using frequency domain decomposition with conditional vector
Year: 2025
Type: conference-paper
Abstract: In this study, we present a bidirectional Urban Safety Network for regional safety management.The proposed Urban Safety Network utilizes artificial neural networks to identify relationships among the structural seismic responses of multiple buildings.To construct a network utilizing Long Short-Term Memory (LSTM), which excels in predicting sequence data, we incorporate not only the time-domain seismic responses of buildings but also conditional vectors derived from frequency domain decomposition as input information.This approach enables the prediction of structural seismic responses that account for the inherent characteristics of buildings under dynamic loads.By using frequency domain decomposition information, the inherent properties of buildings are considered, allowing for the prediction of both linear and nonlinear structural seismic responses by taking into account temporal variations in structural characteristics extracted into the conditional vectors.In the proposed LSTM-based response prediction network, the dynamic structural seismic response of a building assumed to have data loss is intentionally excluded, and only the structural seismic responses of the remaining buildings are used as inputs.The intentionally excluded structural seismic response serves as the output of the LSTM-based model.The LSTM model is trained using a dataset constructed in this manner for all buildings within the target urban area.As a result, the proposed LSTM-based Bidirectional Urban Safety Network can predict and recover lost data by utilizing the structural seismic responses of the remaining buildings, even when data from any individual building within the target set is missing..
Authors: H. Lee, B. Oh
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Funder: Yonsei University
Award: G286147870
OpenAlex ID: W4411488925