Evaluations of the strong ground motion parameter by neural computing and microtremor measurement

Abstract

In this study, a new weight-based neural network model was developed in accordance with a series of historical seismic records to estimate peak ground acceleration at a total of 33 train stations in the Kaohsiung mass rapid transit system of Taiwan. The performance of this model was compared with a simple distribution model and an available ambient vibration survey. The comparison of results showed that the neural network models exhibit a variation tendency similar to the microtremor measurements for all the train stations. The results also showed that over 90% of estimations by the weight-based neural network model were smaller than that of the simple distribution model, and the former model proved to perform better, as the estimations were closer to the survey data for most of the cases. This type of weight-based neural network model might capture the actual response at a construction site more closely, and the results obtained confirm that all train stations comply with the seismic requirement of the building code.