Spatial variation of PGA using neural networks

Abstract

This work is oriented to studying the learning capabilities of neural networks and their impact on the development of a method for the generation of spatial variation of peak ground accelerations PGAs. This spatial variation is based on a limited number of accelerograms registered in specific geotechnical zones, along with a map of soil periods on the Mexico City area. The continuous surfaces that resulted are compared with the most common methods of interpolation, aiming to evidence the substantial advantages of NNs on those of these methods. Based on the results presented, it can be argued that using multi-parameters approaches to define spatial interpolations of a variable as important as PGA, allows to make safer engineering decisions.