Neural networks approach and microtremor measurements in estimating peak ground acceleration due to strong motion

Abstract

Peak ground acceleration is a very important factor that must be considered in construction site for examining the potential damage resulting from earthquake. The actual records by seismometer at stations related to the site may be taken as a basis, but a reliable estimating method may be useful for providing more detailed information of the strong motion characteristics. Therefore, the purpose of this study was by using back-propagation neural networks to develop a model for estimating peak ground acceleration at two main line sections of Kaohsiung Mass Rapid Transit in Taiwan. Additionally, the microtremor measurements with Nakamura transformation technique were taken to further validate the estimations. Three neural networks models with different inputs including epicentral distance, focal depth and magnitude of the earthquake records were trained and the output results were compared with available nonlinear regression analysis. The comparisons exhibited that the present neural networks model did have a better performance than that of the other methods, as the calculation results were more reasonable and closer to the actual seismic records. Besides, the distributions of estimating peak ground acceleration from both of computations and measurements might provide valuable information from theoretical and practical standpoints.