Neural network estimation of ground peak acceleration at stations along Taiwan high-speed rail system

Abstract

It is essential to understand the characteristics of strong motion for reducing the negative impacts in a high-risk area. In this work, a combination of seismic parameters including epicentral distance, focal depth, and magnitude from historical records at 30 checking stations were used in back-propagation neural network model, to estimate peak ground acceleration at ten train stations along the high-speed rail system in Taiwan. The estimation was verified with available microtremor measurement at a specified station, and the calculated horizontal acceleration was checked with the existing building code requirements. A potential hazardous station was identified from the neural network estimation, which exhibited a significantly higher acceleration than that of the design value. The obtained results might be useful for revising the currently applied building code at this region to further fit in the actual earthquake response.