This study introduces a novel framework for complete closed-loop reservoir management (CLRM), integrating automated history matching and production optimization. Two task-specific deep learning models, the improved vision transformer neural network (IViT-NN) and Vision Mamba UNet (VM-UNet), are employed: IViT-NN with ensemble smoother with multiple data assimilation (ES-MDA) for history matching, and VM-UNet with ensemble optimization (EnOpt) for production optimization, where geological uncertainty and varying well controls are taken into account. The proposed approach applied to the three-dimensional Brugge model full-physics simulation requirements by two orders of magnitude and increased the expected net present value (NPV) by approximately 53.9% over a 1,800-day forecast. This provides an efficient approach to real-time dynamic decision-making in reservoir development.
Closed-Loop Reservoir Management Coupling History Matching with Production Optimization Using Deep Learning Surrogates. // Transport in Porous Media. – 2026. – DOI: 10.1007/s11004-026-10357-y. – URL: https://link.springer.com/article/10.1007/s11004-026-10357-y
(2026). Closed-Loop Reservoir Management Coupling History Matching with Production Optimization Using Deep Learning Surrogates. Transport in Porous Media. https://doi.org/10.1007/s11004-026-10357-y
“Closed-Loop Reservoir Management Coupling History Matching with Production Optimization Using Deep Learning Surrogates.” Transport in Porous Media, 2026. doi:10.1007/s11004-026-10357-y.
@article{article35226,
title = {Closed-Loop Reservoir Management Coupling History Matching with Production Optimization Using Deep Learning Surrogates},
journal = {Transport in Porous Media},
year = {2026},
doi = {10.1007/s11004-026-10357-y},
url = {https://link.springer.com/article/10.1007/s11004-026-10357-y}
}
Efficient modeling of subsurface properties with locally varying anisotropy is important for geological characterization, but conventional local-anisotropy simulation can be computationally demanding when large ensembles are required. A...
Regularized inversion returns images that are linear smoothings of an underlying field. The smoothing operator is known but nonstationary. It is the model resolution matrix of the inverse problem. When several such images are available,...
We present a reaction–diffusion (RD) formulation for geostatistical simulation of anisotropic stochastic fields on structured simulation grids. The proposed framework treats stochastic field generation as a variational formulation in...
Due to the inherent complexity of structural geological modeling, it is common for surface-based models to exhibit overlapping surfaces, unrealistic intersections, or simply the need for some kind of editing. These challenges often...
Multivariate geostatistical cosimulation is commonly used to reproduce spatial uncertainty and cross-variable relationships. However, conventional approaches based on simple or ordinary cokriging can become problematic under partially...