سينا ابراهيمي ماكو

The phenomenon of seismic sequences, including foreshock–mainshock an‎d mainshock–aftershock events, is one of the challenging topics in earthquake engineering that plays a crucial role in seismic hazard analysis, as well as in the design an‎d performance of resilient structures. Predicting the seismic response of structures by considering the effects of secondary earthquakes is of fundamental importance in seismic design an‎d disaster management, an‎d remains a major challenge for researchers. In this study, a deep learning–based framework was proposed to simulate seismic sequences, aiming to predict the response spectrum of the secondary earthquake an‎d the Effective Peak Acceleration (EPA) based on the mainshock data. To this end, two hybrid architectures were designed an‎d implemented: the Hybrid CNN–BiLSTM model, combining convolutional layers for local feature extraction with bidirectional LSTMs for capturing temporal dependencies, an‎d the Hybrid TCN model, in which temporal convolutional networks with increasing dilation were used instead of LSTMs to cover both short- an‎d long-term dependencies. The dataset consisted of 84 pairs of sequential earthquakes (mainshock–secondary), provided in three input branches: raw acceleration time history, mainshock response spectrum (Sa), an‎d auxiliary features including period-grouping an‎d peak spectrum characteristics. The outputs of the models included the secondary earthquake spectrum an‎d the EPA value. Model performance was eva‎luated using MAE, RMSE, an‎d R² metrics. The results demonstrated that the CNN–BiLSTM model achieved higher accuracy in predicting both the spectrum an‎d EPA, offering lower MAE an‎d RMSE values compared to the TCN model. In contrast, the TCN model showed relative advantages in training speed an‎d structural simplicity. Overall, the proposed framework provided satisfactory performance in reconstructing secondary earthquake spectra an‎d predicting EPA, an‎d it can serve as a novel tool for seismic hazard analysis an‎d improved seismic design.

پديده توالي لرزه‌اي , مدل CNN–BiLSTM , طيف شتاب , بيشينه شتاب مؤثر , يادگيري عميق

Seismic sequence phenomenon , CNN–BiLSTM model , TCN , response spectrum , Effective Peak Accleration , Deep Learning

Sina Ebrahimi makoo

Dr Ghodrati Amiri,Dr Elham Rajabi