Modeling Lost-Circulation in Fractured Media Using Physics-Based Machine Learning

Albattat, Rami; He, Xupeng; Alsinan, Marwa; Kwak, Hyung; Hoteit, Hussein

doi:10.3997/2214-4609.202210204

Cited by 7 publications

(5 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To deal with this problem, researchers have proposed analytical solutions to predict CO 2 leakage much more efficiently. Avci (1994) proposed an analytical method that combines groundwater flow and the flow through abandoned wells to study the transient flow of the leakage. Nordbotten et al (2004) introduced an analytical solution capable of modeling the leakage in multi-layer and multi-well reservoirs.…”

Section: Introductionmentioning

confidence: 99%

“…He et al (2021b) used the ANN to predict fracture permeability under complex physics, and the ANN surrogate also proved robust on a different dataset collected from the experimental study. Albattat et al (2022) predicted the mud loss using the ANN surrogate, in which the mud loss is related to different physical properties such as the mud rate, bottom hole pressure, and various mud properties. He et al (2022a) predicted the CO 2 storage capacity using the ANN surrogate to reduce computational expense.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Locating CO2 Leakage in Subsurface Traps Using Bayesian Inversion and Deep Learning

Zhang

et al. 2023

Middle East Oil, Gas and Geosciences Show

View full text Add to dashboard Cite

Geologic CO2 sequestration (GCS) is a promising engineering measure to reduce global greenhouse emissions. However, accurate detection of CO2 leakage locations from underground traps remains a challenging problem. This study proposes a workflow that combines Bayesian inversion and deep learning algorithms to detect the sites of CO2 leakage. There are four main steps in the workflow. Step 1: we identify the key uncertainty parameters. Here we mean the CO2 leakage location. Then we get the training set using Latin Hypercube Sampling (LHS) method and perform the high-fidelity simulation using CMG. Step 2: we train the surrogate model using the data set collected from the last step, in which the Bayesian optimization is used to tune the hyperparameters automatically. Step 3: we perform the Bayesian inversion to invert the CO2 leakage location, in which the surrogate serves as the forward model to reduce the computational expense. Step 4: we feed the inverted CO2 leakage location into the high-fidelity model to produce the pressure response. If the error between the pressure response between the surrogate and the high-fidelity model is small enough, the solution is accepted. Otherwise, the accuracy of the surrogate model and the convergence of the Bayesian inversion process are revisited. We validate this method using a synthetic model of CO2 injection. Results show that the proposed Bayesian inversion assisted by the deep learning algorithm can accurately detect the CO2 leakage location with narrow uncertainties. This approach provides an accurate and efficient way to detect CO2 leakage locations in real-time applications.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Locating CO2 Leakage in Subsurface Traps Using Bayesian Inversion and Deep Learning

Zhang

et al. 2023

Middle East Oil, Gas and Geosciences Show

View full text Add to dashboard Cite

show abstract

“…Marzouk et al (2007) deployed Polynomial Chaos Expansion (PCE) surrogate to solve the diffusive problem. Elsheikh et al (2012) utilized Gaussian Process Regression (GPR) to reduce the computational time in a waterflooding reservoir. Naik et al (2019) used the PCE proxy to mimic the reservoir response during polymer flooding.…”

Section: Introductionmentioning

confidence: 99%

“…He et al (2021b) also successfully utilized the ANN surrogate in fracture permeability estimation and CO2 storage prediction (He et al, 2022a). Albattat et al (2022) predicted mud loss using the ANN proxy. As for the CNN application, studies have been done on the upscaling problems in the naturally fractured reservoir to replace the traditional flow-based method (He et al, 2021d;He et al, 2020).…”

Section: Introductionmentioning

confidence: 99%

Parameter Inversion in Geothermal Reservoir Using Markov Chain Monte Carlo and Deep Learning

Zhang

et al. 2023

Day 1 Tue, March 28, 2023

View full text Add to dashboard Cite

Traditional history-matching process suffers from non-uniqueness solutions, subsurface uncertainties, and high computational cost. This work proposes a robust history-matching workflow utilizing the Bayesian Markov Chain Monte Carlo (MCMC) and Bidirectional Long-Short Term Memory (BiLSTM) network to perform history matching under uncertainties for geothermal resource development efficiently. There are mainly four steps. Step 1: Identifying uncertainty parameters. Step 2: The BiLSTM is built to map the nonlinear relationship between the key uncertainty parameters (e.g., injection rates, reservoir temperature, etc.) and time series outputs (temperature of producer). Bayesian optimization is used to automate the tuning process of the hyper-parameters. Step 3: The Bayesian MCMC is performed to inverse the uncertainty parameters. The BiLSTM is served as the forward model to reduce the computational expense. Step 4: If the errors of the predicted response between the high-fidelity model and Bayesian MCMC are high, we need to revisit the accuracy of the BiLSTM and the prior information on the uncertainty parameters. We demonstrate the proposed method using a 3D fractured geothermal reservoir, where the cold water is injected into a geothermal reservoir, and the energy is extracted by producing hot water in a producer. Results show that the proposed Bayesian MCMC and BiLSTM method can successfully inverse the uncertainty parameters with narrow uncertainties by comparing the inversed parameters and the ground truth. We then compare its superiority with models like PCE, Kriging, and SVR, and our method achieves the highest accuracy. We propose a Bayesian MCMC and BiLSTM-based history matching method for uncertainty parameters inversion and demonstrate its accuracy and robustness compared with other models. This approach provides an efficient and practical history-matching method for geothermal extraction with significant uncertainties.

show abstract

“…There are four classes of neural networks, each of which is designed for particular problems. Artificial neural networks (ANNs) specialize in value-to-value applications, including gas injection optimization , mud loss prediction (Albattat et al, 2022), fracture permeability evaluation (He et al, 2021b), etc. ; convolutional neural networks (CNNs) fit image-to-value regression problems like model upscaling (He et al, 2020;Santoso et al, 2019); generative adversarial networks (GANs) focus on image-to-image translation, including image reconstruction Mosser et al, 2017), image super-resolution (Wang et al, 2018;da Wang et al, 2020), etc.…”

Section: Introductionmentioning

confidence: 99%

Rapid NMR T2 Extraction from Micro-CT Images Using Machine Learning

Alsinan

et al. 2022

Day 2 Tue, November 01, 2022

View full text Add to dashboard Cite

Nuclear magnetic resonance (NMR) is an important tool for characterizing pore size distributions of reservoir rocks. Pore-scale simulations from digital rocks (micro-CT images) provide deep insights into the correlation between pore structures and NMR relaxation processes. Conventional NMR simulations using the random walk method could be computationally expensive at high image resolution and particle numbers. This work introduces a novel machine-learning-based approach as an alternative to conventional random walk simulation for rapid estimation of NMR magnetization signals. This work aims to establish a "value-to-value" model using artificial neural networks to create a nonlinear mapping between the input of Minkowski functionals and surface relaxivity, and NMR magnetization signals as the output. The proposed workflow includes three main steps. The first step is to extract subvolumes from digital rock duplicates and characterize their pore geometry using Minkowski functionals. Then random walk simulations are performed to generate the output of the training dataset. An optimized artificial neural network is created using the Bayesian optimization algorithm. Numerical results show that the proposed model, with fewer inputs and simpler network architecture than the referenced model, achieves an excellent prediction accuracy of 99.9% even for the testing dataset. Proper data preprocessing significantly improves training efficiency and accuracy. Moreover, the inputs of the proposed model are more pertinent to NMR relaxation than the referenced model that used twenty-one textural features as input. This works offers an accurate and efficient approach for the rapid estimation of NMR magnetization signals.

show abstract

Modeling Lost-Circulation in Fractured Media Using Physics-Based Machine Learning

Cited by 7 publications

References 0 publications

Locating CO2 Leakage in Subsurface Traps Using Bayesian Inversion and Deep Learning

Locating CO2 Leakage in Subsurface Traps Using Bayesian Inversion and Deep Learning

Parameter Inversion in Geothermal Reservoir Using Markov Chain Monte Carlo and Deep Learning

Rapid NMR T2 Extraction from Micro-CT Images Using Machine Learning

Contact Info

Product

Resources

About