Simulating peak ground acceleration by general regression and radial basis function and other neural networks in some regions of the world

Abstract

Recording of ground motions with high amplitudes of  acceleration and velocity play a key role for designing engineering projects.Here we try to represent a reasonable prediction of peak ground acceleration which may create more than 1g acceleration in different regions.

In this study, applying different structures of Neural Networks and using four key parameters , moment magnitude, rupture distance,  site class , style of faulting which an earthquake may cause serious effects on a site. We introduced a Radial Basis Function Network(RBF) with mean error of 14×10^-3,as the best network for estimating the occurance probability of an earthquake with large value of PGA ?1g in a region. Also the result of applying Back propagation feed forward neural network(FFBP) for predicting high value of PGA, with Mean error of 17×10^-3 , show a good coincidence with the result of the designed RBF Network.

Keywords:Peak ground acceleration, Moment magnitude, Rupture distance, site class, Radial basis function, General regression neural network,.