Real-time Prediction of Tsunami Propagating into the Osaka Bay by Using Artificial Neural Network

Abstract

This study examines the validity to use an artificial neural network (ANN) for the prediction of tsunami magnitudes at several locations in the Osaka Bay by using observed water surface elavations of 20 minumites, at the tower of Shirahama Oceanographic Observatory. The tsunami data used as training and test data for ANN were simulated for different fault models considering tsunami source non-uniformity. The linear activation function was found to be a good choice for output units and the tangent sigmoid function for hidden layer's units. For the training of ANN the Levenberg-Marquardt method with Bayesian regulation were employed. Outputs from the trained network such as the first and the second sea surface falls and rises agreed well with the results of tsunami simulations at each location and all five locations.