poroflow 0.1.0

Physics-informed machine learning for flow in porous media

poroflow

This package is under active development and not yet ready for production use.

Physics-informed machine learning for simulation of flow in porous media.

Overview

poroflow provides methods for simulating multiphase flow in porous media using physics-informed machine learning approaches, including:

• Finite Volume Graph Networks (FVGN) -- GNN-based solvers with built-in conservation guarantees
• Differentiable numerical fluxes -- Godunov flux for correct shock handling
• Progressive rollout training -- autoregressive stability for long-horizon predictions

The initial focus is on the Buckley-Leverett equation for two-phase immiscible displacement, with plans to extend to 2D problems, capillary pressure effects, and parametric surrogates.

Installation

pip install poroflow

Status

This is an early release to reserve the package name. Functional code will be published in upcoming versions.

License

MIT