GANSim: Conditional Facies Simulation Using an Improved Progressive Growing of Generative Adversarial Networks (GANs)

Song, Suihong; Mukerji, Tapan

doi:10.1007/s11004-021-09934-0

Cited by 50 publications

(33 citation statements)

References 23 publications

(16 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…GANs have been used for geomodelling of subsurface reservoirs with either the traditional training approach (Chan & Elsheikh, 2017Dupont et al, 2018;Laloy et al, 2018;Mosser et al, 2020;Nesvold & Mukerji, 2021;Zhang et al, 2019) or the progressive training one (Song et al, 2021a); with whichever way, the generator learns geological patterns from given training facies models. With the learned pattern knowledge, the trained generator can thus produce facies models consistent with the learned patterns, i.e., unconditional geomodelling.…”

Section: Basics Of Gansmentioning

confidence: 99%

“…In recent years, GANs have been combined with geomodelling on different aspects (e.g., Chan & Elsheikh, 2017Dupont et al, 2018;Laloy et al, 2018;Mosser et al, 2020;Zheng & Zhang, 2022;Nesvold & Mukerji, 2021;Song et al, 2021aSong et al, , 2021bSong et al, , 2022Zhang et al, 2019).…”

Section: Introductionmentioning

confidence: 99%

“…For unconditional GANs-based geomodelling, the generator CNN is first trained to abstract spatial geological patterns from training facies models; then it uses the learned pattern knowledge to quickly produce multiple reservoir realizations consistent with the pattern from random latent vectors. Song et al (2021a) proposed to use a progressive training approach for GANs in geomodelling, going incrementally from coarse resolution to finer and finer resolutions and proved that it performs better than the conventional alternative of training where the finest resolution is directly produced.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

GANSim-3D for conditional geomodelling: theory and field application

Song¹,

Mukerji²,

Zhang³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Geomodelling of subsurface reservoirs is important for water resources, hydrocarbon exploitation, and Carbon Capture and Storage (CCS). Traditional geostatistics-based approaches cannot abstract complex geological patterns and are thus not able to simulate very realistic earth models. We present a Generative Adversarial Networks (GANs)-based 3D reservoir simulation framework, GANSim-3D, which can capture geological patterns and relationships between various conditioning data and earth models and is thus able to directly simulate multiple 3D realistic and conditional earth models of arbitrary sizes from given conditioning data. In GANSim-3D, the generator, designed to only include 3D convolutional layers, takes various 3D conditioning data and 3D random latent cubes (composed of random numbers) as inputs and produces a 3D earth model. Two types of losses, the original GANs loss and condition-based loss, are designed to train the generator progressively from shallow to deep layers to learn the geological patterns and relationships from coarse to fine resolutions. Conditioning data can include 3D sparse well facies data, 3D low-resolution probability maps, and global features like facies proportion, channel width, etc. Once trained on a training dataset where each training sample is a 3D cube of a small fixed size, the generator can be used for geomodelling of 3D reservoirs of large arbitrary sizes by directly extending the sizes of all inputs and the output of the generator proportionally. To illustrate how GANSim-3D is used for field geomodelling and also to verify GANSim-3D, a field karst cave reservoir in Tahe area of China is used as an example. The 3D well facies data and 3D probability map of caves obtained from geophysical interpretation are used as conditioning data. First, we create a training dataset consisting of facies models of 64×64×64 cells with a process-mimicking simulation method to integrate field geological patterns. The training well facies data and the training probability map data are produced from the training facies models. Then, the 3D generator is successfully trained and evaluated in two synthetic cases with various metrics. Next, we apply the pretrained generator for conditional geomodelling of two field cave reservoirs of Tahe area. The first reservoir is 800m×800m×64m and is divided into 64×64×64 cells, while the second is 4200m×3200m×96m and is divided into 336×256×96 cells. We fix the input well facies data and cave probability maps and randomly change the input latent cubes to allow the generator to produce multiple diverse cave reservoir realizations, which prove to be consistent with the geological patterns of real Tahe cave reservoir as well as the input conditioning data. The noise in the input probability map is suppressed by the generator. Once trained, the geomodelling process is quite fast: each realization with 336×256×96 cells takes 0.988 seconds using 1 GPU (V100). This study shows that GANSim-3D is robust for fast 3D conditional geomodelling of field reservoirs of arbitrary sizes.

show abstract

Section: Basics Of Gansmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

GANSim-3D for conditional geomodelling: theory and field application

Song¹,

Mukerji²,

Zhang³

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…In Song et al (2021b), condition-based loss functions were used to condition facies on hard data and global features and later they extended the method for spatial probability maps in Song et al (2021a). However, condition-based losses relies on designing manual functions that compute the consistency between the generated samples and target conditions (e.g., computing facies frequency for the generated realizations to mimic real probability maps Song et al (2021a)).…”

Section: Introductionmentioning

confidence: 99%

Generation of non-stationary stochastic fields using Generative Adversarial Networks

Abdellatif¹,

Elsheikh²,

Busby³

et al. 2022

Preprint

View full text Add to dashboard Cite

In the context of generating geological facies conditioned on observed data, samples corresponding to all possible conditions are not generally available in the training set and hence the generation of these realizations depends primary on the generalization capability of the trained generative model. The problem becomes more complex when applied on non-stationary fields. In this work, we investigate the problem of training Generative Adversarial Networks (GANs) models against a dataset of geological channelized patterns that has a few non-stationary spatial modes and examine the training and self-conditioning settings that improve the generalization capability at new spatial modes that were never seen in the given training set. The developed training method allowed for effective learning of the correlation between the spatial conditions (i.e. non-stationary maps) and the realizations implicitly without using additional loss terms or solving a costly optimization problem at the realization generation phase. Our models, trained on real and artificial datasets were able to generate geologically-plausible realizations beyond the training samples with a strong correlation with the target maps.

show abstract

“…A pertinent question then becomes is it possible to use the generator trained on small-size facies models for geomodeling of field reservoirs of large sizes? Third, Song et al (2021aSong et al ( , 2022 only use synthetic cases to validate the proposed GANSim. Given much more sophisticated geological patterns of field reservoirs than synthetic patterns, field reservoirs are needed to verify the effectiveness and efficiency of GANSim.…”

mentioning

confidence: 99%

GANSim‐3D for Conditional Geomodeling: Theory and Field Application

Song

Mukerji

Zhang

et al. 2022

Water Resources Research

Self Cite

View full text Add to dashboard Cite

Geomodeling, or quantitatively predicting the distribution of subsurface reservoirs, is of great significance for evaluation and exploitation of underground water and energy resources as well as for geological sequestration of CO 2 (CCS). Generally, various types of information (data) about the subsurface are incorporated using geostatistical approaches for geomodeling. Such information includes sparse well data, geophysical data, global features (e.g., facies proportion), and spatial geological patterns, among which geological patterns may be the most challenging to incorporate. In traditional geostatistics-based geomodeling approaches, geological patterns can be partially expressed by simple variogram functions (e.g., in the Sequential Indicator Simulation method; Pyrcz & Deutsch, 2014) or local multiple points statistics (MPS) (in MPS-based approaches; Mariethoz & Caers, 2014). Such partial representations may be unable to completely express complicated spatial geological patterns, and thus there is a lack of realism (expected geological patterns) to different extents in the simulated results of these approaches, for example, facies models produced by variogram-based approaches cannot exhibit sinuous channel-like shapes, while MPS-based approaches may produce discontinuous channels. In addition, due to the incompleteness and imperfectness of the incorporated information (e.g., the sparse nature of well data and low-resolution nature of geophysical data), uncertainty exists in geomodeling results, and thus a number of reservoir realizations are generally produced to represent the potential distribution of subsurface reservoirs.Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) in deep learning are effective at capturing complete spatial patterns (structures) of 2D images or 3D objects using a generator Convolutional Neural Network (CNN), with the assistance of another discriminator CNN. With the captured pattern knowledge, the generator CNN can map a random latent vector into a realistic image or a realistic 3D object (e.g., Wu et al., 2016). Text S1 in Supporting Information S1 provides additional details about the basic methodology of GANs.

show abstract

GANSim: Conditional Facies Simulation Using an Improved Progressive Growing of Generative Adversarial Networks (GANs)

Cited by 50 publications

References 23 publications

GANSim-3D for conditional geomodelling: theory and field application

GANSim-3D for conditional geomodelling: theory and field application

Generation of non-stationary stochastic fields using Generative Adversarial Networks

GANSim‐3D for Conditional Geomodeling: Theory and Field Application

Contact Info

Product

Resources

About