Prediction of large peak ground acceleration (PGA- exceeded the value of gravity acceleration) would play a key role in earthquake disaster reduction projects. In this study, we apply two methods of artificial neural network (ANN) and support vector machine (SVM) to predict some earthquakes with PGA larger than 1g. We have proposed general regression neural network (GRNN) and multi layer perceptron (MLP) with back propagation algorithm for ANN method. Input data consists of earthquake magnitude, rupture length, fault mechanism and surface geology of the area, related to 60 events had occurred all over the world for the period of 1971 to 2004. 26 events out of 60 earthquakes have PGA≥1 (Strasser and Bommer, 2009). We present optimum ANN and SVM models for the data set and show that both models have good ability to predict PGA, especially GRNN model for better prediction and compatibility with observational data.

Keywords