Fast Marching Method-Inspired Inversion Neural Network for Permeability Prediction
<p id="p1">Characterizing permeability heterogeneity is crucial for accurate reservoir modeling, as it dominates fluid flow in subsurface reservoirs. Traditional methods for permeability estimation typically couple reservoir simulators with iterative inversion methods. In addition, the estimation of...
Saved in:
| Main Author: | |
|---|---|
| Format: | Book Chapter |
| Language: | English |
| Published: |
IntechOpen
2025
|
| Subjects: | |
| Online Access: | https://www.intechopen.com/chapters/89814 |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | <p id="p1">Characterizing permeability heterogeneity is crucial for accurate reservoir modeling, as it dominates fluid flow in subsurface reservoirs. Traditional methods for permeability estimation typically couple reservoir simulators with iterative inversion methods. In addition, the estimation of heterogeneous permeability fields can be challenging because of the high dimensionality of the geological models, monitoring data paucity, and high computational costs of simulations. All these factors may lead to high uncertainty for fluid flow behavior predictions. This chapter introduces recent deep learning models for permeability estimation inspired by the fast marching method (FMM). These models can predict permeability fields from pressure derivative data. Such “data” are obtained using a semi-analytic asymptotic solution to the diffusivity equation that uses the diffusive time of flight (DTOF), which itself can be efficiently calculated by using the FMM. Due to the spatial nature of the permeability estimation, deep neural networks (DNNs) are adopted to perform the inversion. The first inversion neural network (INN) model takes the pressure derivative collected at sparse observational locations as input variables and inversely estimates the permeability field. Further, an ensemble INN (EINN) is introduced to directly learn the nonlinear mapping between the innovation vector and the update vector in the history-matching problem. With the ensemble feature, the EINN can effectively quantify the uncertainty in the heterogeneous permeability fields. The applicability of the INN and EINN in permeability estimation is demonstrated with reasonable accuracy and high efficiency, which represents a significant leap forward in the inversion methods for reservoir engineering, geoscience, and hydrogeology.</p> |
|---|---|
| ISBN: | 978-1-83968-501-9 |
| DOI: | 10.5772/intechopen.115629 |