第一部門：物理数学力学問題 （B）

[S01B-01] Unknown parameter estimation using Physics Informed Neural Networks with noised observation data

*Shota Deguchi¹, Yosuke Shibata¹, Mitsuteru Asai¹ (1. Kyushu University)

Keywords:Deep Learning, Neural Networks, Inverse Problem

With the 2011 off the Pacific coast of Tohoku Earthquake and the late frequency increase in heavy rainfall, numerical simulations have often been employed for the disaster prediction. However, city-scale numerical simulations sometimes require high-resolution models and large computational costs. One can replace those models with equivalent-permeability-having porous media in order for the computational efficiency. In this study, instead of traditional man-powered operations or empirical laws, parameter selection of the media is interpreted as an inverse problem with the help of Physics-Informed Neural Networks (PINNs). Since limited data observations could be done in general engineering issues, this study particularly surveys the inverse problem applicability of PINNs.