Investigating steady unconfined groundwater flow using Physics Informed Neural Networks

A deep learning technique called Physics Informed Neural Networks (PINNs) is adapted to study steady groundwater flow in unconfined aquifers. This utilizes information from underlying physics represented the form of partial differential equations (PDEs) alongside data obtained physical observations. In this work, we consider Dupuit–Boussinesq equation, which based on Dupuit–Forchheimer approximation, as well a recent, more complete model derived by Di Nucci (2018) models. We then train PINNs steady-state analytical solutions and laboratory experiments. Using PINNs, predict phreatic surface profiles given different input conditions recover estimates for hydraulic conductivity experimental show that can eliminate inherent inability equation flows with seepage faces. Moreover, inclusion models constrains solution space produces better predictions than training alone. are robust little effect added noise data. Furthermore, compare via examine effects higher order terms included formulation but neglected approximation. Lastly, discuss effectiveness using examining flow.