Inverse PINN Identification of Viscosity Ratio in Buckley–Leverett Transport

  • Imankulov, Timur (Al-Farabi Kazakh National University)
  • Kenzhebek, Yerzhan (Al-Farabi Kazakh National University)
  • Bekele, Samson Dawit (Al-Farabi Kazakh National University)
  • Akhmed-Zaki, Darkhan (Mukhtar Auezov South Kazakhstan University)

Please login to view abstract download link

Reliable management of subsurface energy systems requires estimating flow parameters that are difficult to measure directly. The viscosity ratio Mstrongly controls displacement efficiency and the propagation speed of saturation fronts in Buckley–Leverett transport [1]. We develop an inverse physics-informed neural network (PINN) that reconstructs the saturation field S(x,t) while identifying M from sparse, noisy saturation observations [2], targeting history-matching settings where discontinuities are present. The forward model is the hyperbolic conservation law ∂S/∂t+∂f(S;M)/∂x=0, where f(S;M)is the fractional-flow function parameterized by quadratic relative permeabilities. A multilayer perceptron S_θ (x,t) represents saturation; a sigmoid output enforces the physical bounds 0