NEURAL-NETWORK BASED ESTIMATION OF NORMALIZED RESPONSE SPECTRA

摘要

This paper focuses on the application of neural networks as an alternative computational tool for the estimation of normalized response spectra for the horizontal ground motions with magnitudes M_(JMA) ≥ 5 and hypocentral distances less than 50 km. The feasibility of using the perceptron neural networks in estimating site-specific response spectra and the effects of the geophysical properties of the site is examined. Two neural-network models are proposed for generating normalized response spectra, such that those consider the effects of local site conditions. Model 1 is developed with six inputs (i.e., magnitude, hypocentral distance, primary wave velocity, shear wave velocity, N-values obtained by the standard penetration test (SPT), and density of soil), whereas Model 2 is developed with three inputs (i.e., magnitude, hypocentral distance, and shear wave velocity). As expected, a better performance is obtained from (neural-network) Model 1 in terms of accuracy and efficiency. The results obtained from this study are very encouraging and have a potential to replace the commonly used regression approach.