Sustainable Materials For Transitional and Alternative Energy

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/357061435

Minimizing Carbon Footprint by Smart Sustainable Reservoir Management

Conference Paper · December 2021


DOI: 10.2118/204752-MS

CITATIONS READS
0 190

4 authors, including:

Klemens Katterbauer A. Marsala


EUCLID 85 PUBLICATIONS   700 CITATIONS   
99 PUBLICATIONS   223 CITATIONS   
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Klemens Katterbauer on 21 December 2021.

The user has requested enhancement of the downloaded file.


SPE-204752-MS

Minimizing Carbon Footprint by Smart Sustainable Reservoir Management

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


Klemens Katterbauer, Alberto Marsala, Abdulaziz Al Qasim, and Ali Yousif, Saudi Aramco

Copyright 2021, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Middle East Oil & Gas Show and Conference, Manama, Bahrain, 28 November – 1 December, 2021. The event
was cancelled. The official proceedings were published online on 15 December, 2021.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Sustainability and reducing carbon footprint has attracted attention in the oil and gas industry to optimize
recovery and increase efficiency. The 4th Industrial Revolution has made an enormous impact in the oil
and gas industry and on analyzing carbon footprint reduction opportunities. This allows classification of
various reservoir operations, installation of permanent sensors and robots on the field, and reduction of
overall power consumption.
We present an overview of new AI approaches for optimizing reservoir performance while reducing their
carbon footprint. We will outline the significant carbon emissions contributors for field operations and how
their impact will change throughout the production's lifecycle from a reservoir. Based on this analysis, we
will outline via an AI-driven optimization framework areas of improvement to reduce the carbon footprint
considering the uncertainty.
We analyzed the framework's performance on a synthetic reservoir model with several producing wells,
water, and CO2 injecting wells. Beneficial in reducing carbon emissions from the field is the reuse and
injection of CO2 for enhancing hydrocarbon production from the reservoir. One hundred different scenarios
were then investigated utilizing an innovative autoregressive network model to determine the impact of these
components on the overall carbon emission of the field and determine its uncertainty. The conclusions from
the analysis were then incorporated into a data-driven optimization routine to minimize carbon footprint
while maximizing reservoir performance. The final optimization results of the showcase outlined the ability
to reduce the carbon footprint significantly.

Introduction
Sustainability in the oil and gas sector has grown in importance in the last several decades to extend the
life-cycle of reservoirs. The challenges of reducing the carbon footprint while at the same time maintaining
production levels have even become more challenging with the primary drive mechanism needing to be
supplemented by the injection of water to maintain production levels [1]. In recent decades, there have been
several initiatives to minimize the carbon footprint of reservoir operations while maximizing recovery from
these fields. CO2 sequestration has become one of the most promising initiatives for dealing with rising
carbon emissions in the atmosphere and simultaneously enhancing recovery from the reservoir [2, 3, 4].
2 SPE-204752-MS

For the challenge of optimizing CO2-assisted recovery optimization while at the same time reducing the
carbon footprint of the operations, conventional approaches are to develop a reservoir model and history
match the reservoir to the existing production and injection data [5]. While such an approach allows accurate
modeling of the reservoir structure, the computational time required to run a simulation makes an automatic
optimization approach with several hundreds of simulations an infeasibility. The operational impact factors
of carbon emissions still have to be taken into account. Data-driven approaches utilizing machine learning
and artificial intelligence have the advantage of capturing the dynamics in the well production data while

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


being efficient to be executed [6, 7]. A significant advantage for data-driven approaches is that biases
incorporated by false subsurface modeling or interpretations affecting significantly reservoir simulation
results are avoided, and the algorithms solely base their learning on the collected data [8].
Conventional approaches, such as the use of history-matched reservoir models, have been deployed to
optimize CO2-assisted recovery, while at the same time reducing the carbon footprint of the operations [5,
9]. While such approaches allow to take into account the detailed reservoir formation structure, there are
several uncertain parameters that may considerably affect the forecast or may bias the reservoir forecasting
results. Specifically, aquifer strength, reservoir fracture network and permeability as well as well as fault
connectivity may be highly uncertain [4]. While history matching aims to estimate these parameters
and assess the uncertainty, the resulting inversion problem is highly ill-posed with a large number of
different solutions. Furthermore, the computational time required to run a simulation makes it practically not
feasible to deploy an automatic optimization approach with several hundreds of simulation scenarios [10].
Furthermore, the impact of operational factors (such as drilling of new wells or changes in well schedules) on
carbon emissions must still be considered and modeled. Data-driven approaches utilizing machine learning
and artificial intelligence have the advantage of capturing the production dynamics, while being efficient for
execution over hundreds of scenarios. Another important advantage for data-driven approaches is that biases
introduced by false subsurface modeling or interpretations that can significantly affect reservoir simulation
results are avoided. This is due to the fact that the algorithms solely incorporate the information inherent in
the data, without any additional bias caused by interpreted data.
Data driven AI approaches have found wider adoption in recent years for more accurately estimating
reservoir properties far from wellbores. Ertekin et al. provided an extensive overview of the applications
of artificial intelligence in reservoir engineering problems [11]. The authors present an overview about
advanced machine learning approaches and more state-of-the-art artificial intelligence technologies for the
reservoir engineering domain. A particular focus in the article is on advanced machine learning algorithms
and their utilization for addressing challenges encountered in the extraction of hydrocarbons. Fuzzy logic
and response surface model approaches are widely deployed for the categorization of reservoir zones and
characterization of the properties within these zones [12]. Lim et. al. presented a fuzzy logic approach
for determining the porosity and permeability distribution in the reservoir from well logs. The results of
combining the Fuzzy Curve approach together with a neural network approach allows implement a nonlinear
regression approach in determining the field's permeability and porosity properties. While the approach
is quite promising, the authors solely focused on the well log parameter estimation, and did not attempt
to estimate permeability fields. In another article, David Wood presents a nearest-neighborhood approach
for the joint estimation of porosity, permeability and saturation in well log data [13]. The utilized TOB
algorithm exhibited strong performance for the estimation of the well logs, however, it may face challenges
due to overfitting when trying to predict reservoir saturation maps. A different approach by Tariq et al.
utilizes a functional network model for the estimation of water saturation in carbonate rocks based on well
logs [14]. While the approach delivers promising results for the data under investigation, the nonlinearity in
saturation that is encountered in fractured carbonate rocks may make the approach challenging to be utilized
for field saturation estimation.
Extensive research has been conducted in trying to estimate water saturation from well logs and several
algorithms have been utilized to better estimate water saturation profiles from well logs in the near wellbore
SPE-204752-MS 3

area. While these approaches allow to have an accurate understanding of the saturation profile in the near
wellbore profile, the estimates may be far off in the interwell region given the heterogeneity of the rock
formation and lack of interwell direct measurements.
Deep electromagnetic tomography has become one of the viable solutions to have a more accurate
understanding of the interwell reservoir volume, leveraging on the resistivity contrast between water and
hydrocarbons [15]. Additionally, crosswell electromagnetic tomography also has been a promising solution
for enhancing reservoir history matching and fluid saturation mapping via complementing well logs and

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


production information with interwell volumetric resistivity data [16, 17].
Several crosswell electromagnetic (EM) surveys were conducted in recent years, demonstrating the
significant potential of this technology for interwell saturation mapping. A crosswell electromagnetic survey
was performed between two horizontal wells, a watered out oil producer and a water injector in the reservoir
section [18], followed by a 3D resistivity inversion that was developed to derive a 3D resistivity cube in the
interwell space [19]. The main strength of this unique technology is how it offers an incomparable depth of
investigation. It also provides a robust way to assist with the extraction of the water saturation distribution
[20].
Key challenge to obtain an accurate fluid distribution mapping using an analytical approach (i.e., Archie's
equation) from a tomographic 3D resistivity cube, up to few kilometers in size, is the lack of knowledge
of the controlling parameters, such as water salinity, cementation, factor, saturation exponent, porosity and
fracture distribution, etc., in the interwell volume. The problem is mathematically ill-posed, as we do not
have direct methods to measure the volumetric distribution of such quantities in the interwell reservoir
volumes. Different approaches were undertaken to tackle this problem, such as using advanced uncertainty
analysis techniques [21], integration with history matching workflows or data assimilation methods.
Developments in deep learning have profoundly impacted the oil and gas industry, allowing accurate
analysis and interpretation of reservoir data [22]. Deep learning and machine learning concern computers’
ability to learn without being explicitly programmed and represents a core area of modern computational
techniques. Machine learning can generally be classified into three major categories that distinguish
themselves in terms of the response signal available to them. The three major categories are supervised
and unsupervised learning and reinforcement learning [23]. Supervised learning is the broadest category
and implies that both input and target data are available, consisting of numerical values or string labels,
and the responses are known. This allows us to compare the responses to the target labels and learn from
the difference to improve the models. Unsupervised learning learns from answers without having available
any associated responses. The algorithms cluster the data to associated features and provide deeper analysis
and correlations between the data. The challenge of unsupervised learning is that the feature extraction
may depend strongly on the underlying algorithm, given the lack of ground truth. Reinforcement learning
is a major area for improving unsupervised learning by providing positive or negative feedback for the
algorithms’ decisions [24]. This approach is similar to trial and error and penalizes the estimates for errors.
For example, this may be the utilization of different strategies for the choke size opening to optimize
production from reservoirs. Recently, semi-supervised learning has attracted significant attention as there
are ways to utilize incomplete training data where some of the target outputs are missing. Machine learning
problems are differentiated into classification, regression, and clustering approaches [25, 6]. Classification
requires that the inputs are subdivided into different classes, and the classes have to be known in advance.
Regression problems focus on the continuous estimate of outputs from inputs, and clustering requires the
division of input into separate groups. While the problem may be formulated in various ways depending on
the objective, posing the estimation of the reservoir injection parameters as a continuous regression problem
is quite commonly performed.
For determining and forecasting reservoir production based on injection and associated reservoir
parameters, a time-series regression approach represents a viable approach given the ability to take into
4 SPE-204752-MS

account the temporal changes in the reservoir production, while incorporating reservoir depletion derived
changes in the production performance [26, 27].
In the next chapters, we present a novel artificial intelligence framework for a data-driven approach to
enhance reservoir production while minimizing the carbon footprint of the operations.

Methodology
The developed advanced deep learning framework for carbon footprint water and gas injection optimization

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


consists of a nonlinear autoregressive network connected to a nonlinear continuous optimization framework
for the minimization of the overall carbon footprint of the reservoir while maximizing production. As
displayed in Figure 1, the framework takes the historical reservoir data, in particular production and injection
rates from the wells and preprocesses them. The preprocessing step involves the removal of any outliers and
filtering of sharp spikes in the data which indicate measurement errors. It then incorporates them into an
autoregressive network to estimate the total oil production from the producer wells in addition to the carbon
footprint of the entire reservoirs. The arising estimates and trained deep learning nonlinear autoregressive
network are then utilized to optimize the injection quantities for water and CO2. The objective of the
optimization framework is to maximize production while minimizing at the same time the carbon footprint
of the operations.

Results
We examined the framework on a synthetic reservoir model (Figure 2) and outlined the wells’ training data
for the 10 years of simulation for water and CO2 injection in Figure 3. The reservoir consists of four water
injection wells, three CO2 injectors and six producer wells. We also outline the corresponding production
patterns for the six different production wells related to oil, water, and gas production. As can be seen, the
production pattern is somewhat fluctuating, indicating a trend for increased gas production across all the
producer wells compared to a reduction in water production and oil production.

Figure 1—Carbon footprint minimization framework incorporating a deep learning nonlinear autoregressive
network for the optimization of production and minimization of the carbon footprint of the reservoir.
SPE-204752-MS 5

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


Figure 2—Synthetic 3D reservoir model with six producers (green),
and four water injectors (blue) and three CO2 injectors (red).

Figure 3—Training data sets for the synthetic reservoir model.


6 SPE-204752-MS

The first step in developing the deep learning framework was the development of the nonlinear
autoregressive network model. We trained the model on the reservoir injection and production rates,
assuming a delay in the input state of half a month between the injection and production correlation. For
this, we take into account for each estimation the three last days of the injection rates. A comparison of the
estimates compared to the total oil production (Figure 4) outlines high accuracy estimation performance
of the trend of the production levels for the training data. The same is the case for the estimation of the
overall carbon footprint. The results indicate the model's good performance to forecast the production from

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


injection levels for the reservoir model under consideration. This yields high confidence in the model used
in the subsequent carbon footprint optimization framework for smart reservoir management.

Figure 4—Comparison of the estimates versus the target training.

Based on the trained framework, we then examined the framework to forecast production levels for
the next five and a half years and displayed the optimized injection patterns in Figure 5. The results
demonstrate relatively stable water injection quantities for each of the four injector wells, varying in terms
of the maximum injection volume by only around 10 percentage points. The CO2 injection patterns show a
different behavior with a peak in the injection after around 3.5 years for the well I5CO2, whereas the other
two CO2 injector wells exhibit a decreasing injection trend.
More importantly are the results in Figure 6, which outline the optimized cumulative oil production
and cumulative carbon footprint of the reservoir. A comparison between the base and optimal scenario
exhibits the considerable improvement in oil recovery, while simultaneously significantly reducing the
carbon footprint over the lifespan of the reservoir. Oil recovery over the 5.5 year time span was increased
by 12.57% as compared to the base case, while the carbon footprint was reduced by more than 65% for
the entire reservoir.
SPE-204752-MS 7

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


Figure 5—Outline of the optimized water and CO2 injection profiles.

Figure 6—Comparison of the production forecast levels for 100 different scenarios.

Conclusion
We have developed an innovative nonlinear autoregressive deep learning framework to minimize the
carbon footprint of reservoirs while maintaining production levels. The framework incorporates a nonlinear
autoregressive deep network model that was then integrated into a nonlinear optimization framework. The
results outlined the framework's strong performance to minimize carbon emissions from the field while
maximizing the oil recovery. Additionally, the deep learning model for forecasting production levels from
the reservoir outlined the capability to deliver accurate forecasts based on the various injection strategies.

References
[1] K. Bennaceur, "How the Oil and Gas Industry Is Contributing to Sustainability.," Journal of
Petroleum Technology, vol. 71, no. 3, pp. 38–39, 2019.
[2] M. Şükrü, "Analysis of the effect of experimental adsorption uncertainty on CH4 production and
CO2 sequestration in Dadas shale gas reservoir by numerical simulations.," Journal of Petroleum
Science and Engineering, pp. 1051–1066, 2019.
8 SPE-204752-MS

[3] A. Hassan, S. Elkatatny and A. Abdulraheem, "Intelligent prediction of minimum miscibility


pressure (MMP) during CO2 flooding using artificial intelligence techniques.," Sustainability,
vol. 11, no. 24, p. 7020, 2019.
[4] K. Katterbauer, I. Hoteit and S. Sun, "History Matching of Electromagnetically Heated
Reservoirs Incorporating Full-Wavefield Seismic and Electromagnetic Imaging," SPE Journal,
vol. 20, no. 5, pp. 923–941, 2015.
[5] Z. Dai, R. Middleton, H. Viswanathan, J. Fessenden-Rahn, J. Bauman, R. Pawar, S.-Y. Lee and

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


B. McPherson, "An integrated framework for optimizing CO2 sequestration and enhanced oil
recovery.," Environmental Science & Technology Letters, pp. 49–54, 2014.
[6] K. Katterbauer and A. Marsala, "A novel 4IR framework for interwell saturation mapping," in
82nd EAGE Annual Conference & Exhibition, 2020.
[7] Y. Zhang, M. M. Hittawe, K. Katterbauer, A. F. Marsala, O. M. Knio and I. Hoteit, "Joint seismic
and electromagnetic inversion for reservoir mapping using a deep learning aided feature-oriented
approach," in SEG Technical Program Expanded Abstracts 2020, Virtual, 2020.
[8] K. Katterbauer, A. Marsala, M. Maucec, Y. Zhang and I. Hoteit, "History Matching of Time-
Lapse Deep Electromagnetic Tomography with A Feature Oriented Ensemble-Based Approach,"
in ECMOR XVII, Edingburgh, 2020.
[9] K. Katterbauer, A. Marsala, V. Schoepf and E. Donzier, "A novel artificial intelligence automatic
detection framework to increase reliability of PLT gas bubble sensing," Journal of Petroleum
Exploration and Production, pp. 1–11, 2021.
[10] F. Sana, K. Katterbauer, T. Al-Naffouri and I. Hoteit, "Orthogonal matching pursuit for
enhanced recovery of sparse geological structures with the ensemble Kalman filter," IEEE
Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 9, no. 4,
pp. 1710–1724, 2016.
[11] T. Ertekin and Q. Sun, "Artificial Intelligence Applications in Reservoir Engineering: A Status
Check," energies, 2019.
[12] J.-S. Lim, "Reservoir properties determination using fuzzy logic and neural networks from
well data in offshore Korea," Journal of Petroleum Science and Engineering, vol. 49, no. 3-4,
pp. 182–192, 2005.
[13] D. Wood, "Predicting porosity, permeability and water saturation applying an optimized
nearest-neighbour, machine-learning and data-mining network of well-log data," Journal of
Petroleum Science and Engineering, vol. 184, 2020.
[14] Z. Tariq, M. Mahmoud and A. Abdulraheem, "An Artificial Intelligence Approach to Predict
the Water Saturation inCarbonate Reservoir Rocks," in SPE-195804-MS, SPE Annual
Technical Conference and Exhibition, Calgary, 2019.
[15] A. Marsala and S. Ma, "Crosswell Electromagnetic Tomography: from Resistivity Mapping
to interwell Fluid Distribution," in IPTC 12229 presented at the International Petroleum
Technology Conference, Kuala Lumpur, 2008.
[16] K. Katterbauer, I. Hoteit and S. Sun, "Synergizing Crosswell Seismic and Electromagnetic
Techniques for Enhancing Reservoir Characterization," SPE-174559-PA - SPE Journal, vol.
21, no. 3, 2016.
[17] K. Katterbauer, I. Hoteit and S. Sun, "A Time Domain Update Method for Reservoir History
Matching of Electromagnetic Data," in Offshore Technology Conference, 2014.
[18] A. Marsala and S. Lyngra, "Crosswell Electromagnetic Induction between Two Widely
Spaced Horizontal Wells: Coiled-Tubing Conveyed Data Collection and 3D Inversion from
a Carbonate Reservoir in Saudi Arabia," in SEG International Exposition and 85th Annual
Meeting, New Orleans, 2015.
SPE-204752-MS 9

[19] A. Marsala, S. Lyngra and S. AlSaif, "3D Inversion Practice for Crosswell Electromagnetic
Surveys in Horizontal Wells in Saudi Arabia," in SEG International Exposition and 85th
Annual Meeting, New Orleans, 2015.
[20] A. Marsala, S. Lyngra, M. Ma and P. Zhang, "Workflow to Integrate Geophysical Deep
EM and Reservoir Simulation for Inter-Well Saturation Mapping presented," in 79th EAGE
Conference & Exhibition, Paris, 2017.
[21] S. Saif and A. Marsala, "Uncertainty Quantification Algorithms for Reservoir Characterization

Downloaded from http://onepetro.org/SPEMEOS/proceedings-pdf/21MEOS/4-21MEOS/D041S038R006/2547619/spe-204752-ms.pdf/1 by Saudi Aramco user on 21 December 2021


of Interwell Volumes," in AAPG International Conference and Exhibition, Capte Town, 2018.
[22] J. You, W. Ampomah and Q. Sun, "Development and application of a machine learning
based multi-objective optimization workflow for CO2-EOR projects.," Fuel, vol. 264, p.
116758116758, 2020.
[23] S. Wang and S. Chen, "Insights to fracture stimulation design in unconventional reservoirs
based on machine learning modeling.," Journal of Petroleum Science and Engineering, pp.
682–695, 2019.
[24] K. Katterbauer and A. Marsala, "A Novel Sparsity Deploying Reinforcement Deep Learning
Algorithm for Saturation Mapping of Oil and Gas Reservoirs," Arabian Journal for Science
and Engineering, pp. 1–7, 2020.
[25] Y. Ao, H. Li, L. Zhu, S. Ali and Z. Yang, "The linear random forest algorithm and its
advantages in machine learning assisted logging regression modeling.," Journal of Petroleum
Science and Engineering, vol. 174, pp. 776–789, 2019.
[26] H. Jeong, A. Sun, J. Lee and B. Min, "A learning-based data-driven forecast approach for
predicting future reservoir performance," Advances in Water Resources, pp. 95–109, 2018.
[27] S. Turner, K. Doering and N. Voisin, "Data-Driven Reservoir Simulation in a Large-Scale
Hydrological and Water Resource Model," Water Resources Research, 2020.
[28] K. Katterbauer, A. Al-Yousif and A. Marsala, "Intelligent Reconciliation of Well Logs–A
Pathway Towards 4IR Assisted Log Interpretation," in Abu Dhabi International Petroleum
Exhibition & Conference., 2020.
[29] K. Katterbauer and A. Marsala, "A Novel 4IR Framework for Interwell Saturation Mapping,"
in 82nd EAGE Annual Conference & Exhibition, Amsterdam, 2020.

View publication stats

You might also like