DEVELOPMENT OF PROBABILISTIC RISK EVALUATION METHOD WITH SURROGATE-MODELED NUMERICAL SIMULATIONS

Abstract

With Nankai Trough Earthquake expected to strike in the near future while natural disasters are getting intensified lately, their risk evaluations are urgently being called for. Since it is, in general, very challenging to provide precise damage estimations to natural phenomena, which are often accompanied by countless uncertainties, probabilistic approaches are commonly adopted to account for them. Nowadays, numerical simulations are widely applied for disaster predictions. Although they support high accuracy in terms of anticipating, their computational costs per unit calculation are extremely large, and hence, not suitable for developing probabilistic discussions in practice. This study focuses on a singular value decomposition-based surrogate model, which expresses the numerical analysis approximately. This paper made a modification on the surrogate model by newly introducing kernel ridge regression. The accuracy and effectiveness of the proposed method were examined through an inundation simulation and a tsunami simulation.