Aug 2026· Water Resources Research· Vol 62· 0 citations· 40 references
TL;DR
This study proposes an end‐to‐end inversion modeling workflow, termed DDPM‐DSI, which integrates a Denoising Diffusion Probabilistic Model (DDPM) with Data‐Space Inversion (DSI), and indicates that DDPM‐DSI provides a promising generative inversion framework for efficient history matching and geological parameter estimation under complex geological conditions.
Abstract
Traditional geological parameter inversion methods often require repeated forward simulations for history matching, leading to high computational cost and limited inversion efficiency. Their performance may also be restricted when geological fields exhibit strong non‐Gaussian characteristics. To address these challenges, this study proposes an end‐to‐end inversion modeling workflow, termed DDPM‐DSI, which integrates a Denoising Diffusion Probabilistic Model (DDPM) with Data‐Space Inversion (DSI). In the proposed workflow, a probabilistic sparse attention encoder is used to extract temporal features from multivariate production data and provide conditioning information for the diffusion model. Geological parameter fields are represented in a reduced latent space using Principal Component Analysis (PCA) for Gaussian‐like cases and a Vector‐Quantized Variational Autoencoder (VQ‐VAE) for channelized non‐Gaussian cases. The DSI method is then applied to generate posterior dynamic‐response samples from observed data and prior simulations, which are used to condition the trained diffusion model for direct generation of posterior geological parameter fields. The proposed workflow is evaluated using a 2D synthetic reservoir model, a 3D Brugge benchmark model, and a non‐Gaussian Egg benchmark model. The results show that DDPM‐DSI can recover major geological structures and reproduce production‐response trends in the tested cases. Compared with traditional GAN‐based inversion, DDPM‐DSI generally produces sharper and more geologically consistent posterior realizations, particularly for non‐Gaussian channelized fields. Compared with the Ensemble Smoother with Multiple Data Assimilation (ESMDA), DDPM‐DSI provides rapid online inference after problem‐specific training and avoids repeated simulator‐based parameter updates during inversion. These results indicate that DDPM‐DSI provides a promising generative inversion framework for efficient history matching and geological parameter estimation under complex geological conditions.
Geophysical inversion is inherently non-unique, and efficient uncertainty quantification remains a major challenge for large-scale applications. Conventional Bayesian approaches, such as Markov chain Monte Carlo (McMC), provide rigorous posterior estimates but are often computationally prohibitive, while deterministi...
Inversions of dynamic multi‐scale multi‐physics subsurface system properties rely on the effective sampling of high‐dimensional parameter spaces and are thus notoriously difficult to perform. The associated models are defined by complex nonlinear relationships with potentially stochastic components, which lead to ana...
Kwabena Atobra, G. Hillers, V. Lyakhovsky et al.· Journal of Geophysical Resea...· 0 citations
Full waveform inversion (FWI) reconstructs subsurface velocity models from seismic mea surements but faces the challenges of ill-posedness and nonlinearity, making accurate inversion dependent on suitable initialization and effective prior constraints. Classical methods stabi lize the inversion and improve convergence...
Chen Min, Hao-Wen Jiang, Zheng Ma et al.· arXiv.org· 0 citations
Data-driven site characterization in geotechnical engineering increasingly relies on high-dimensional waveform data and computationally intensive inverse modeling. Full waveform inversion and finite element model updating typically rely on gradient-based or Bayesian optimization, requiring many serial forward simulatio...
Anthony LoRe Starleaf, S. Parida, Souvik Chakraborty et al.· Geotechnics· 0 citations
Solutions to geophysical inverse problems are generally non-unique, and greatly benefit from additional geological prior information. Generative Neural Networks (GNNs) are commonly used to parametrise such information, typically restricting solutions to a lower-dimensional latent space. The final inversion result is ob...
Pratyay Roy, Xue-Bin Zhao, Andrew Curtis· 0 citations
Recent advances in deep‐learning‐based earthquake detection have revolutionized seismology, producing high‐resolution catalogs complete down to very low magnitudes. While these data sets offer an unprecedented view of seismic triggering, their massive scale and inherent short‐term incompleteness pose a major challenge...
P. Bountzis, G. Petrillo, E. Lippiello· Journal of Geophysical Resea...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.