10:45 〜 11:05
[AGE27-01] Physics-Informed Neural Networksを用いた土壌水分シミュレーション
★招待講演
Water flow in soils can be described by the Richardson-Richards equation (RRE), which is a nonlinear partial differential equation. Because its analytical solution is not available in practical situations, it has been solved by numerical methods, such as the finite difference method, finite element method, and finite volume method. We here introduce an alternative numerical method called physics-informed neural networks (PINNs), where the solution to the RRE is approximated by neural networks based on their universal approximation capability. Although PINNs require more computational resources to solve the forward problem of the RRE compared to the other numerical methods, the PINNs approach is expected to be effective for the inverse problem. This is because the PINNs approach does not need repetitive solutions of the forward problem as in other numerical methods. Here, we introduce the PINNs solver for the RRE, and its solution was compared with the analytical and other numerical solutions to the RRE for homogeneous and layered soils. To demonstrate the potential of the PINNs approach for the inverse problem, the PINNs approach was applied to the estimation of surface flux from near-surface soil moisture measurements.