Temporal convolution neural network model for simulation of site seismic effect

As a hotspot and difficulty in geotechnical earthquake engineering, site seismic effect simulation is mainly studied by mathematical and physical methods or based on observational records, facing challenges such as solving dynamic equations, modeling uncertainty, data sparsity and generalization ability. In this study, a physics-embedded temporal convolution neural network （Phy-TCN） model is considered to verify the performance difference compared with the data-driven temporal convolution neural network （TCN）. The Phy-TCN model is used to simulate site seismic effect according to the on-site/borehole strong motion records in KiK-net database. The results show that the Phy-TCN model can effectively simulate time series data. In the simulation of signals containing noise from KiK-net observation records, the average relative errors of the Phy-TCN and equivalent linearization method are 0.067 and 0.379, respectively, based on the response spectra values at specific periodic points of earthquake events at selected sites. In conclusion, Phy-TCN model can be applied to the simulation of site seismic effect under the condition of fuzzy soil profile.