Ground Motion Prediction Model Using Artificial Neural Network

This article focuses on developing a ground motion prediction equation based on artificial neural network (ANN) technique for shallow crustal earthquakes. A hybrid technique combining genetic algorithm and Levenberg–Marquardt technique is used for training the model. The present model is developed to predict peak ground velocity, and 5% damped spectral acceleration. The input parameters for the prediction are moment magnitude (Mw), closest distance to rupture plane (Rrup), shear wave velocity in the region (Vs30) and focal mechanism (F). A total of 13,552 ground motion records from 288 earthquakes provided by the updated NGA-West2 database released by Pacific Engineering Research Center are utilized to develop the model. The ANN architecture considered for the model consists of 192 unknowns including weights and biases of all the interconnected nodes. The performance of the model is observed to be within the prescribed error limits. In addition, the results from the study are found to be comparable with the existing relations in the global database. The developed model is further demonstrated by estimating site-specific response spectra for Shimla city located in Himalayan region.