长短期频域特征嵌入的建筑结构地震响应预测

A reliable seismic response prediction model can effectively predict the dynamic responses of building structures under earthquake excitations, thereby reducing disaster losses and safeguarding lives and property. In this study, a data-driven model, FreResGRU, which integrated time-domain and frequency-domain features, was proposed to predict the dynamic responses of building structures under seismic excitations. The FreResGRU model represents ground motion acceleration in the complex domain. This approach enables an in-depth learning of both the macroscopic periodic evolution and the local transient characteristics of acceleration sequences, thereby more comprehensively extracting the key features of amplitude-frequency nonlinear coupling and dynamic temporal correlations in seismic acceleration signals. Furthermore, three numerically simulated datasets and one dataset based on real earthquake events were employed to comprehensively evaluate the predictive performance of the FreResGRU model. The results show that the predictive performance of the FreResGRU model is superior to that of existing models, such as LSTM-s, ResLSTM, Pyramid-LSTM, Pyramid-Transformer, Py-GA. Compared with the second-best model, the FreResGRU model reduces the root mean square error by an average of 31.75%.