Structural Response Estimation Using Gated Recurrent Unit

Abstract

Recent tragedies have demonstrated that natural disasters, such as earthquakes and typhoons, can wreak havoc on society. Numerical models and simulations are used for predicting the structural response and damage caused by disasters. However, some structures do not have any design drawings or numerical models, and thus, problems are encountered when conducting numerical simulations. Furthermore, even if the model exists, the response predicted through numerical simulation may be different from the response of the actual structure. Although effort has been made to resolve this issue using model-updating techniques, these methods are laborious for developing a new model that reflects the current state of the structure. Therefore, the aim of this study is to develop a new method that automatically predicts the time-series response of structures using a deep learning technique. The gated recurrent unit), based on the recurrent neural network, was used to predict the structural response. Simulation-based validation tests were conducted to verify the performance of the proposed method. The proposed method could estimate the response of the structure with a root-mean-square error of 13.59%.