Surrogate models for Rock-Fluid Interaction: A Grid-Size-Invariant Approach

2026-02-25 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences · Depth: Expert, quick

Summary

This work introduces eight surrogate models designed to predict fluid flow in porous media, addressing the high computational cost of conventional high-fidelity numerical models for rock-fluid interaction. Four models are reduced-order models (ROMs) utilizing two neural networks for compression and prediction. The other four are single neural networks featuring grid-size invariance, meaning they can infer on computational domains larger than their training data. The study compares UNet and UNet++ architectures, demonstrating that UNet++ offers superior predictive performance for these surrogate models. The grid-size-invariant approach is shown to reliably reduce memory consumption during training, yielding strong correlation with ground-truth values and outperforming the ROMs analyzed, even for challenging scenarios like fluid-induced rock dissolution.

Key takeaway

For AI Scientists developing models for multi-query problems in porous media, consider implementing grid-size-invariant neural networks. This approach significantly reduces memory consumption during training and allows for inference on larger computational domains, making it highly suitable for uncertainty quantification and optimization tasks where numerous scenarios must be evaluated efficiently. Prioritize UNet++ architectures over UNet for superior predictive performance.

Key insights

Grid-size-invariant neural networks offer efficient, scalable surrogate modeling for complex rock-fluid interactions.

Principles

• UNet++ outperforms UNet for surrogate models.
• Grid-size invariance reduces training memory consumption.

Method

Develop surrogate models using either reduced-order models (two neural networks) or single grid-size-invariant neural networks capable of inferring on larger domains than trained.

In practice

• Employ UNet++ for improved surrogate model performance.
• Use grid-size-invariant models for memory-efficient training.

Topics

• Surrogate Models
• Rock-Fluid Interaction
• Grid-Size Invariance
• UNet++ Architecture
• Reduced-Order Models

Best for: AI Scientist, AI Researcher, Research Scientist, Machine Learning Engineer

Related on AIssential

• Toward AI-Driven Digital Twins for Metropolitan Floods: A Conditional Latent Dynamics Network Surrogate of the Shallow Water Equations — CLDNet significantly accelerates metropolitan flood forecasting by using a latent neural ODE and coordinate-based decoding.
• An Improved Generative Adversarial Network for Micro-Resistivity Imaging Logging Restoration — A specialized GAN architecture significantly improves micro-resistivity imaging log restoration by integrating multi-scale feature learning and dual discrimination.
• Structure-Aware Epistemic Uncertainty Quantification for Neural Operator PDE Surrogates — Restricting uncertainty sampling to the lifting module improves neural operator uncertainty quantification efficiency and accuracy.
• Separable Neural Architectures as Physical World Models: from Mathematical Theory to Applications — The SNA combines neural networks with tensor decomposition to efficiently model high-dimensional PDEs, enabling rapid simulations and inversions.
• The Reservoir Attention Network: Cross-Pass State in Pretrained Transformers via Content-Addressable Reservoir Injection — The Reservoir Attention Network enables cross-pass state in pretrained transformers using an untrained, fixed reservoir.
• Deep Learning Surrogates for Emulating Stochastic Climate Tipping Dynamics — A dynamics-informed Temporal Fusion Transformer provides a 465x faster, differentiable surrogate for forecasting stochastic climate tipping events with high fidelity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.