LAViG-FLOW: Latent Autoregressive Video Generation for Fluid Flow Simulations

TL;DR

LAViG-FLOW is a novel latent autoregressive diffusion model that efficiently generates realistic, time-consistent fluid flow fields for subsurface applications, significantly reducing computational costs compared to traditional simulators.

Contribution

It introduces a new diffusion-based framework with autoencoders and transformers for coupled pressure and saturation field prediction in fluid flow simulations.

Findings

Generates consistent flow fields faster than traditional methods

Learns coupled pressure and saturation dynamics effectively

Achieves two orders of magnitude speedup in simulations

Abstract

Modeling and forecasting subsurface multiphase fluid flow fields underpin applications ranging from geological CO2 sequestration (GCS) operations to geothermal production. This is essential for ensuring both operational performance and long-term safety. While high fidelity multiphase simulators are widely used for this purpose, they become prohibitively expensive once many forward runs are required for inversion purposes and to quantify uncertainty. To tackle this challenge, we propose LAViG-FLOW, a latent autoregressive video generation diffusion framework that explicitly learns the coupled evolution of saturation and pressure fields. Each state variable is compressed by a dedicated 2D autoencoder, and a Video Diffusion Transformer (VDiT) models their coupled distribution across time. We first train the model on a given time horizon to learn their coupled relationship and then fine-tune it autoregressively so it can extrapolate beyond the observed time window. Evaluated on an open-source CO2 sequestration dataset, LAViG-FLOW generates saturation and pressure fields that stay consistent across time while running two orders of magnitude faster than traditional numerical solvers.