Seismic Ground Response Prediction Based on Multilayer Perceptron

Earthquake disasters can cause enormous social and economic damage, therefore the sustainability of infrastructure requires mitigation earthquake consequences. In seismic design infrastructures, it is essential to estimate response site during earthquake. Geotechnical engineers have developed quantitative methods for analyzing ground response. This study proposes a multilayer perceptron (MLP) model evaluate surface based on motion at bedrock (or 100 m level), compares its performance with that conventional model. A total 6 sites, events each site, were selected from Kiban Kyoshin Network (KiK-net) used as datasets. The acceleration spectra calculated predicted measured (baseline) histories compared. proposed MLP magnitudes natural periods where amplifies closely motions (baseline). outperformed analysis. However, did not perform well earthquakes whose exceed 2 g due deficiency in large measurements training