Peak Ground Acceleration Prediction Using Artificial Neural Networks Approach: Application to the Kik-Net Data

Abstract

The aim of this work is to propose a prediction equation of the PGA using the Multi-Layer Perceptron Artificial Neural Network method (ANN) with a Levenberg–Marquardt backpropagation algorithm for the training. The inputs are the magnitude, the focal depth, the epicentral distance, the thickness and the mean frequency at up to a shear wave velocity equal to 800m/s, while the target result is the PGA. To establish this network, data collected from the KIK-NET seismic data base in Japan area have been used. 102 sites and 1850 records are used in the training phase while 326 events were not used in the training are kept for the test phase. The obtained results show that PGA computed using the ANN method are close to those recorded. Finally, a example is presented, in which, 55 records are used to compared the ANN method with two Ground Motion Prediction Equations (GMPEs). This example demonstrates how the ANN is more robust than classical models.