applied

sciences
Article
Machine Learning-Assisted Prediction of Oil Production and
CO2 Storage Effect in CO2-Water-Alternating-Gas
Injection (CO2-WAG)
Hangyu Li 1,2 , Changping Gong 1,2 , Shuyang Liu 1,2, * , Jianchun Xu 1,2 and Gloire Imani 1,2

1 Key Laboratory of Unconventional Oil & Gas Development, China University of Petroleum (East China),
Ministry of Education, Qingdao 266580, China
2 School of Petroleum Engineering, China University of Petroleum (East China), Qingdao 266580, China
* Correspondence: [email protected]

Abstract: In recent years, CO2 flooding has emerged as an efficient method for improving oil recovery.
It also has the advantage of storing CO2 underground. As one of the promising types of CO2 enhanced
oil recovery (CO2 -EOR), CO2 water-alternating-gas injection (CO2 -WAG) can suppress CO2 fingering
and early breakthrough problems that occur during oil recovery by CO2 flooding. However, the
evaluation of CO2 -WAG is strongly dependent on the injection parameters, which in turn renders
numerical simulations computationally expensive. So, in this work, machine learning is used to help
predict how well CO2 -WAG will work when different injection parameters are used. A total of 216
models were built by using CMG numerical simulation software to represent CO2 -WAG development
scenarios of various injection parameters where 70% of them were used as training sets and 30% as
testing sets. A random forest regression algorithm was used to predict CO2 -WAG performance in
terms of oil production, CO2 storage amount, and CO2 storage efficiency. The CO2 -WAG period,
CO2 injection rate, and water–gas ratio were chosen as the three main characteristics of injection
parameters. The prediction results showed that the predicted value of the test set was very close to
Citation: Li, H.; Gong, C.; Liu, S.; Xu, the true value. The average absolute prediction deviations of cumulative oil production, CO2 storage
J.; Imani, G. Machine Learning- amount, and CO2 storage efficiency were 1.10%, 3.04%, and 2.24%, respectively. Furthermore, it only
Assisted Prediction of Oil Production takes about 10 s to predict the results of all 216 scenarios by using machine learning methods, while the
and CO2 Storage Effect in CO2 -Water- CMG simulation method spends about 108 min. It demonstrated that the proposed machine-learning
Alternating-Gas Injection (CO2 -WAG).
method can rapidly predict CO2 -WAG performance with high accuracy and high computational
Appl. Sci. 2022, 12, 10958. https://
efficiency under conditions of various injection parameters. This work gives more insights into the
doi.org/10.3390/app122110958
optimization of the injection parameters for CO2 -EOR.
Academic Editor: Flavio Cannavò
Keywords: CO2 -WAG; machine learning; injection parameters; random forest regression; CCUS
Received: 12 October 2022
Accepted: 26 October 2022
Published: 29 October 2022

Publisher’s Note: MDPI stays neutral 1. Introduction

with regard to jurisdictional claims in
With the continuous improvement of oil and gas exploration and development, the pro-
published maps and institutional affil-
portion of heterogeneous and low-permeability reservoirs in exploration and development
iations.
is gradually increasing. Using conventional water injection to develop low-permeability
reservoirs results in a low recovery factor [1]. Compared to water, CO2 is less viscous
and can enter into small pores more easily, which can reduce the viscosity of crude oil,
Copyright: © 2022 by the authors. expand the volume of crude oil, improve the mobility ratio, and thus increase the oil
Licensee MDPI, Basel, Switzerland. recovery factor [2,3]. Therefore, CO2 -enhanced oil recovery (CO2 -EOR) has great potential
This article is an open access article for developing low-permeability reservoirs. In addition to this, CO2 -EOR can sequestrate
distributed under the terms and CO2 underground, which can help in achieving carbon neutrality [4]. However, in the CO2
conditions of the Creative Commons flooding method, during gas injection, the low viscosity of CO2 may lead to a phenomenon
Attribution (CC BY) license (https:// referred to as viscous fingering. This results in an unfavorable mobility ratio, which seri-
creativecommons.org/licenses/by/ ously affects the improvement of swept volume [5]. Moreover, due to its low density, CO2
4.0/).

Appl. Sci. 2022, 12, 10958. https://doi.org/10.3390/app122110958 https://www.mdpi.com/journal/applsci

Appl. Sci. 2022, 12, 10958 2 of 18

can easily escape to the upper part of the reservoir, forming a fugitive flow channel [6]. This
also constitutes a drawback for CO2 flooding as it leads to a reduction of the swept volume.
In response to the problems of waterflooding and CO2 flooding, researchers proposed
CO2 -WAG methods [7]. This technology was first used by Mobil in 1957 in a sandstone
reservoir in Alberta, Canada. It combines the characteristics of water flooding and CO2
flooding, which increases the macroscopic sweep efficiency and thus improves the overall
oil displacement efficiency. In addition, CO2 -WAG can mitigate the issue of rapid CO2 flow
and increase the gas phase’s flow resistance. It can also lower the resistance to the flow of
the water phase and increase the mobility ratio. As a result of this, recovery efficiency can
be greatly improved [7]. It has been reported that 80% of oilfield projects in the US using
WAG technology have achieved good results [8]. Skauge et al. [9] studied 59 WAG fields
and found that the average recovery of crude oil was improved by 10% for all WAG cases.
Recently, WAG has also been used in the Brazilian subsalt oilfield complex [10]. Subsalt
oil and gas production was 2739 million barrels of oil equivalent per day (2.739 Mboe/day)
by 2020, representing 70.3% of Brazil’s total oil equivalent. The Lula field, which conducted
the WAG pilot test in April 2011, has a cumulative oil production of 2000 Mboe by 2020 and
is the largest extracted/producing field in Brazil, with an average oil and gas production of
988,000 barrels/day and 43.2 Mm3 /day, respectively [11].
Currently, the optimization of WAG extraction schemes is the focus of many oil
fields and related researchers [12–20]. Rodrigues et al. [21] used CMG reservoir numerical
simulation software to optimize the application of WAG in a sub-salt offshore field in Brazil
and proposed a design method for CO2 -WAG operations in carbonate reservoirs, focusing
on the economics, the CO2 cycle efficiency, and project risk. It is worth mentioning that
the application of intelligent algorithms such as machine learning, which has developed
rapidly in recent years, has been used in petroleum exploration and development [22,23],
especially in optimization problems. For instance, Bilgesu et al. [24] proposed a method for
bit optimization with the help of neural networks. Leite Cristofaro and Longhin et al. [25]
optimized the mud loss problem in Brazilian deepwater subsalt fields with the help of
KNN, MLP, and NB algorithms. Wang et al. [26] proposed a joint optimization method for
well location and injection and extraction parameters using the random forest as well as a
radial basis neural network. In general, memory-based learning algorithms perform better
than any other family of algorithms. These methods assume that a given set of terms and
class labels can be used as a mapping to identify unlabeled term classes [25].
Random forest is a decision tree-based machine learning algorithm proposed by
Breiman and Cutler in 2001 [27]. The random forest regression model is built by com-
bining the results obtained from several well-established decision tree models, and the
final prediction result is obtained by averaging the prediction results of all decision tree
models [28]. A large number of studies [29–32] have shown that random forest models
have the advantages of strong generalization ability, insensitivity to input data deviations,
and the ability to analyze the importance of input features. In this study, by combining the
random regression forest algorithm with the numerical simulations, a method for rapidly
forecasting the cumulative oil production, CO2 storage amount, and CO2 storage efficiency
of CO2 -WAG development schemes has been developed. This method can significantly
increase the effectiveness of scheme optimization in oilfields.

2. Methods
2.1. CMG Base Model
The simulations are carried out by using a simulator known as the Computer Mod-
eling Group Ltd. (CMG). The submodule GEM of the CMG simulator is a compositional
simulator and it is widely employed for simulating the displacement behavior of CO2 -EOR
in reservoir formations. Thus, this work employed the submodule GEM of CMG to conduct
the simulation of CO2 -WAG.
Appl. Sci. 2022, 12, x FOR PEER REVIEW 3 of 19

Appl. Sci. 2022, 12, 10958 3 of 18

EOR in reservoir formations. Thus, this work employed the submodule GEM of CMG to
conduct the simulation of CO2-WAG.

2.1.1. Parameter
2.1.1. Parameter Settings
Settings ofof the
the Base
Base Model
Model
AA five-spot
five-spotwell
wellpattern
patternis is established
established using
using CMG
CMG software,
software, as shown
as shown in Figure
in Figure 1.
1. The
The model
model has ahas
wella in
well
thein the middle
middle wherewhere
water water
and CO and CObe
2 can 2 can be injected
injected alternatively.
alternatively. There
There
are are 23both
23 grids gridsinboth
the in the XYand
X and Y directions,
directions, with awith
gridasize
gridof size × 30
30 of 30 ××303 ×m3inmthe
in the X,
X, Y,
and Z directions,
Y, and Z directions,respectively.
respectively. The model
The modelis is
divided
dividedinto
into8 8layers theZZdirection,
layersininthe direction,and
and
thus
thus the
the total
total number
number of of grids
grids is 23 ×
is 23 23 ××88==4232.
× 23 4232.

Figure 1. Three-deminsional
Figure 1. Three-deminsional diagram
diagramof
ofthe
theWater
WaterAlternating
AlternatingGas
Gasmodel.
model.

The
The key
key reservoir
reservoir parameters
parameters used
used in
in this
this work
work are
are selected
selected based
based onon the
the geological
geological
information
information of the Tuo 28 block in Shengli Oilfield [33,34], as shown in Table1.1. The
of the Tuo 28 block in Shengli Oilfield [33,34], as shown in Table The oil
oil
reservoir depth is 1800 m, and the reservoir pressure is 18 MPa due to the normal
reservoir depth is 1800 m, and the reservoir pressure is 18 ◦MPa due to the normal for- formation
pressure coefficient
mation pressure of 1.0. The
coefficient reservoir
of 1.0. temperature
The reservoir is 85 is
temperature C.85The
°C.porosity is 0.24,
The porosity the
is 0.24,
initial oil saturation is 0.7934, the crude oil viscosity is 15.4495 cp, and the crude oil density
the initial oil saturation is 0.7934, the crude oil viscosity is 15.4495 cp, and the crude oil
is 760.9 kg/m3 .
density is 760.9 kg/m3.
Table 1. Key parameters of the model.
Table 1. Key parameters of the model.
Serial No. Input Parameters Unit Value
Serial No. Input Parameters Unit Value
11 Average permeability in the X and Y directions − 3
×10 ×10µm 3 5050
Average permeability in the X and Y directions −3 μm3
2 Average permeability in the Z direction × 10−3 µm 3 5
2 Average permeability in the Z direction ×10 μm
−3 3 5
3 Oil layer thickness m 27
43 Oil layer
Top depth thickness m m 180027
54 Top depth
Porosity 1 m 1800
0.24
65 Initial oilPorosity
saturation 1 1 0.7934
0.24
76 Crude
Initialoiloilviscosity
saturation cp 1 15.4495
0.7934
8 Crude oil density kg/m3 760.9
9
7 Crude oil viscosity
Reservoir temperature ◦ C cp 15.4495
85
8
10 InitialCrude
reservoiroil pressure
density MPa kg/m3 760.9
18
9 Reservoir temperature °C 85
10 permeability in the
The Initial reservoirdirections,
horizontal pressure for each layerMPa from top to 18bottom,
is 10 × 10−3 µm3 , 20 × 10−3 µm3 , 30 × 10−3 µm3 , 40 × 10−3 µm3 , 60 × 10−3 µm3 ,
70 × The
10−3permeability
µm3 and 90 in × the
10−horizontal
3 µm3 , as directions, for each
shown in Table layerthe
2, and from top topermeability
average bottom, is 10
× 10 −3 μm3, 20 × 10−3 μm3, 30 × 10−3 μm
− 3 3, 403× 10−3 μm3, 60 × 10−3 μm3, 70 × 10−3 μm3 and 90 ×
in horizontal directions is 50 × 10 µm . The permeability in the vertical direction is
10−3times
0.1 μm3, the
as shown in Table
horizontal 2, and the(K
permeability average permeability in horizontal directions is 50
v /K = 0.1) and the average permeability in the
h
vertical direction is 5 × 10−3 µm3 , which is a typical non-homogeneous low permeability
reservoir with a positive rhythm. The thickness of the whole reservoir is 27 m, with each
layer measuring 3 m.
 × 10−3 μm3. The permeability in the vertical direction is 0.1 times the horizontal permeabil-
ity (Kv/Kh = 0.1) and the average permeability in the vertical direction is 5 × 10−3 μm3, which
is a typical non-homogeneous low permeability reservoir with a positive rhythm. The
Appl. Sci. 2022, 12, 10958 4 of 18
thickness of the whole reservoir is 27 m, with each layer measuring 3 m.

Table 2. Permeability in horizontal and vertical directions.

Table 2. Permeability in horizontal and vertical directions.
Permeability (×10−3 μm3)
Layer −3 µm3 ) Direction
Permeability (×10Vertical
Horizontal Direction
Layer
1 10
Horizontal Direction Vertical 1Direction
2 1 20 10 21
3 2 30 20 32
4 3 40 30 43
40
5 4
60 64
5 60 6
6 6 70 70 77
7 7 80 80 88
8 8 90 90 99

2.1.2. Injection and Production Settings

The model simulates
simulates a total of forty years, with depletion development occurring
during
during thethefirst
firsteight
eightyears
yearsand andCO CO 2-WAG
2 -WAG production
production commencing
commencing in the ninth.
in the Thus,
ninth. the
Thus,
actual CO2CO
the actual -WAG2 -WAGdevelopment
development length is
length 32isyears.
32 In
years. this
In case,
this to
case, compare
to comparethe production
the produc-
tion effects
effects of each
of each CO2-WAGCO2 -WAG scenario,
scenario, it is necessary
it is necessary to carryto out
carry out a period
a period of depletion
of depletion devel-
development
opment first tofirst to exclude
exclude the natural
the natural energy energy disturbance.
disturbance. As shown Asinshown
Figurein2,Figure 2, the
the pressure
pressure difference between the injection and production wells
difference between the injection and production wells is less than 0.5 MPa after the first is less than 0.5 MPa after
8
the first
years of8depletion
years of depletion
development. development.
In addition, In addition, the production
the production efficiency
efficiency is lowis low
andandno
no longer
longer productive.
productive. Therefore,
Therefore, it is
it is necessary
necessary totoadd
addenergy,
energy,such suchas as through
through CO22-WAG-WAG
development. In the base base model,
model,CO CO22 injection
injectionstarts
startsininJanuary
Januaryofofthe theninth
ninthyear
yearand
andwater
wa-
injection
ter starts
injection in July,
starts so the
in July, cycle
so the is set
cycle to 12
is set to months,
12 months, including
including 6 months
6 months of gas injection
of gas injec-
and and
tion 6 months
6 monthsof water injection.
of water The The
injection. water–gas
water–gas ratioratio
is setis to
set1:1, which
to 1:1, implies
which thatthat
implies the
CO injection rate is equal to the water injection rate under reservoir
the 2CO2 injection rate is equal to the water injection rate under reservoir conditions. The conditions. The water
injection rate is rate
set at 3 m3 /day.
water injection is 90
setmat /day
90 m3and /daythe andCO 2 injection
the rate israte
CO2 injection set isatset
20,000
at 20,000 The
m3/day.
bottom
The holehole
bottom pressure
pressureof the fourfour
of the production
production wells is set
wells to 7toMPa.
is set 7 MPa. AsAs shown
shown inin
Figure
Figure3,
the water cut in the produced fluid is more than 80% by the
3, the water cut in the produced fluid is more than 80% by the fortieth year of production. fortieth year of production.
Thus, this
Thus, this work
work assumes
assumes that that production
production is is stopped
stopped after
after 4040 years
years of of production.
production.

Figure 2.
Figure Bottom-hole pressure
2. Bottom-hole pressure in
in injection
injection and
and production
production wells.
wells.
Appl. Sci.Sci.
Appl. 2022, 12, 12,
2022, x FOR PEER REVIEW
10958 5 of 519
of 18

Figure
Figure 3. Oil
3. Oil cutcut
andand water
water cutcut of the
of the produced
produced fluid
fluid in the
in the standard
standard condition.
condition.

2.2. Machine Learning for CO2 -WAG Prediction

2.2. Machine Learning for CO2-WAG Prediction
2.2.1. Principle of the Random Forest Regression Algorithm
2.2.1. Principle of the Random Forest Regression Algorithm
Multiple regression decision trees constitute the random forest regression algorithm.
Multiple
Based on theregression decision trees
idea of integrated constitute
learning, the value
the mean random forestregression
of each regressiondecision
algorithm.
tree is
Based on the idea of integrated learning,
taken as the prediction result [35]. the mean value of each regression decision tree
is taken as the prediction result [35].
1T T
T t∑
 ( )
f ( x ) =1 { f ( x, θt )} (1)
f ( x) = =1 f x, t (1)
T t =1
where: f ( x ) is the model prediction result, f ( x, θt ) is the output based on x and θt , x is the
where: f ( x) isvariable,
independent the modelθt isprediction (
result, identically
the independent )
f x,t distributed
is the output on x and
basedvector,
random andT is

 t , x Asisathe
the number of regression decision trees.
independent
machine learningvariable,
algorithm
 t based
is the independent
on statistical identically
theory, thedistributed
random forest ran- re-
domgression and T is
vector,algorithm introduces
the number theofbagging method
regression and the
decision random subspace method [36]
trees.
to As
avoid the problem
a machine learning of single
algorithmdecision
basedtree models, which
on statistical theory,tend
the to be overfitted
random forest re-and
not accurate.
gression algorithm introduces the bagging method and the random subspace method [36]
to avoid(1) The
the bagging
problem of method [37], also
single decision treeknown
models, as which
bootstrap
tendaggregating,
to be overfitted is a and
bootstrap-
not
based statistical method. Based on repeatable random sampling, multiple predictors are
accurate.
formed
(1) The bybagging
the bootstrap
method repetitive sampling
[37], also known method. Assuming
as bootstrap that there
aggregating, is aare N samples
bootstrap-
in the original sample, N samples are repeatedly sampled to
based statistical method. Based on repeatable random sampling, multiple predictorsform new training samples.
are
formed by the bootstrap repetitive sampling method. Assuming that there are N samples is
When N approaches infinity, the probability of not being sampled again for each sample
36.8%.
in the Nearly
original 36.8% N
sample, of samples
the original samples will
are repeatedly not appear
sampled in the
to form newtraining
trainingsamples of the
samples.
When N approaches infinity, the probability of not being sampled again for each sampleThe
same tree, and the samples that are not drawn are called out-of-bag data (OOB) [27].
generation
is 36.8%. of locally
Nearly 36.8% of optimal solutions
the original for regression
samples decisionintrees
will not appear can be avoided
the training samplesbyofthe
bagging method.
the same tree, and the samples that are not drawn are called out-of-bag data (OOB) [27].
(2) Stochastic
The generation subspace
of locally optimal method.
solutions Random featuresdecision
for regression need totrees
be selected when con-
can be avoided
structing the regression
by the bagging method. decision tree. Selecting random features means picking x feature
attributes at random from the whole set of attributes. Node splitting
(2) Stochastic subspace method. Random features need to be selected when construct- selects the optimal
features based on the principle of minimum mean squared deviation
ing the regression decision tree. Selecting random features means picking x feature attrib- so that each tree is
not pruned to achieve maximum growth. A random sampling of training samples and a
utes at random from the whole set of attributes. Node splitting selects the optimal features
random selection of feature attributes can make sure that the regression decision trees have
based on the principle of minimum mean squared deviation so that each tree is not pruned
as much variety as possible [38].
to achieve maximum growth. A random sampling of training samples and a random se-
lection of feature attributes can make sure that the regression decision trees have as much
variety as possible [38].
Appl. Sci. 2022, 12, x FOR PEER REVIEW 6 of 19
Appl. Sci. 2022, 12, 10958 6 of 18

2.2.2. Randomized Regression Forest Algorithm Flow

2.2.2. Randomized Regression Forest Algorithm Flow
The main flow of the randomized regression forest algorithm is presented in Figure
The main flow of the randomized regression forest algorithm is presented in Figure 4
4 and described as follows:
and described as follows:
(1) Sampling: K sets of datasets are sampled from the training dataset S by using the
(1) Sampling: K sets of datasets are sampled from the training dataset S by using the
bagging method. Each set of datasets is divided into 2 types: sampled data and un-sam-
bagging method. Each set of datasets is divided into 2 types: sampled data and un-sampled
pled data (out-of-bag data), and they are trained to produce a decision tree.
data (out-of-bag data), and they are trained to produce a decision tree.
(2) Growing: Each decision tree is trained by training data. In each branching node,
(2) Growing: Each decision tree is trained by training data. In each branching node,
M features are
M features arerandomly
randomlyselected
selectedfrom
from MM feature attributes,
feature andand
attributes, the the
bestbest
features are cho-
features are
sen based on the Gini index for full branching growth until no more growth
chosen based on the Gini index for full branching growth until no more growth is possible is possible
without pruning.
without pruning.
(3) Forming aa forest:
(3) Forming forest: Repeat
Repeat steps
steps 11 and
and 22 to
to build
build multiple
multiple regression
regression decision
decision trees
trees
and maximize the growth of each tree to form
and maximize the growth of each tree to form a forest. a forest.
(4) Predicting:
(4) Predicting: Using
Using the
the chosen
chosen model,
model, predictions
predictions are made about
are made about the
the new
new data
data set,
set,
and the final output is the average of all the predictions made by the decision
and the final output is the average of all the predictions made by the decision trees. trees.

Figure 4. The
Figure 4. The training
training process
process of
of the
the random
random forest
forest algorithm.
algorithm.

This
This work
workfirst
firstbuilds
buildsaaseries
seriesofofnumerical
numerical simulation
simulationmodels
modelsby by
adjusting thethe
adjusting injection
injec-
parameters and runs them to obtain simulation results for forming a test database. The
tion parameters and runs them to obtain simulation results for forming a test database.
train_test_split function from sklearn.model_selection is called to randomly divide the
The train_test_split function from sklearn.model_selection is called to randomly divide
test database into two parts, one as a training set to train the regression prediction model
the test database into two parts, one as a training set to train the regression prediction
and the other as a test set to compare with the results predicted by machine learning to
model and the other as a test set to compare with the results predicted by machine learn-
verify the accuracy of the method. To make the results clearer and more intuitive, this
ing to verify the accuracy of the method. To make the results clearer and more intuitive,
work trains and predicts each of the three label variables separately with the random forest
this work trains and predicts each of the three label variables separately with the random
regression algorithm.
forest regression algorithm.
The CO2 -WAG period, fluid injection rate, and water–gas ratio are important pa-
The CO2-WAG period, fluid injection rate, and water–gas ratio are important param-
rameters for CO2 -WAG optimization. Therefore, this work uses CO2 -WAG period, CO2
eters for CO2-WAG optimization. Therefore, this work uses CO2-WAG period, CO2 injec-
injection rate and water–gas ratio as three features and cumulative oil production, CO2
tion rate and water–gas ratio as three features and cumulative oil production, CO2 storage
storage amount, and CO2 storage efficiency as labels for regression prediction to realize
amount,
fast and CO
prediction storage efficiency
of 2program effects withasthe
labels
helpfor
ofregression prediction
a random forest to realize
regression fast pre-
algorithm in
diction of program
machine learning. effects with the help of a random forest regression algorithm in ma-
chine learning.
3. Results and Discussion
3. Results
3.1. andAnalysis
Base Case Discussion
3.1. Base
The Case Analysis
cumulative oil production, CO2 storage amount, and CO2 storage efficiency of the
base model as a function
The cumulative oil of time are shown
production, in Figures
CO2 storage 5 and 6.
amount, TheCO
and values of cumulative
2 storage efficiencyoil
of
production, CO storage amount, and CO
the base model2as a function of time are shown2 storage efficiency are derived as the three
in Figures 5 and 6. The values of cumula- labels
for
tivemachine learning.CO
oil production, The CO2 storage
2 storage amount
amount, is obtained
and CO 2 storageby subtracting
efficiency are the cumulative
derived as the
CO 2 production
three from the learning.
labels for machine cumulative CO
The CO2 injection.
2 storage The CO2 is
amount storage efficiency
obtained is expressed
by subtracting the
as the ratio of
cumulative COthe CO2 storage
2 production amount
from to the CO2CO
the cumulative injection.
2 injection. The CO2 storage efficiency

is expressed as the ratio of the CO2 storage amount to the CO2 injection.
 As shown in Figure 5, at first, the oil production rate goes up quickly because there
is enough energy in the reservoir. As the reservoir energy gradually depletes, the oil pro-
duction curve begins to level off gradually in the fifth year. Then, the reservoir energy is
replenished by the nineth year after the CO2-WAG started, and the oil production rate
Appl. Sci. 2022, 12, 10958 begins to increase rapidly. The oil production rate begins to decrease slowly from7 the of 18
twentieth year to the end of production.

Figure 5. Cumulative oil production in the production process.

At the beginning of the CO2 injection, the CO2 storage amount is almost equal to the
CO2 injection amount from the nineth year to the fifteenth year with little CO2 produced
(Figure 6a). CO2 production begins in the fifteenth year and has a relatively stable rate.
From the sixteenth year, the growth rate of CO2 storage starts to be lower than that of CO2
injection. With the increase in CO2 production, the CO2 storage amount nearly stopped
increasing by the thirty-sixth year. Consequently, the CO2 storage efficiency is almost
equal to5.5.
Figure
Figure 1Cumulative
at the beginning
Cumulative and then in
oilproduction
oil productiongradually
inthe decreases,
theproduction
production as shown in Figure 6b.
process.
process.

At the beginning of the CO2 injection, the CO2 storage amount is almost equal to the
CO2 injection amount from the nineth year to the fifteenth year with little CO2 produced
(Figure 6a). CO2 production begins in the fifteenth year and has a relatively stable rate.
From the sixteenth year, the growth rate of CO2 storage starts to be lower than that of CO2
injection. With the increase in CO2 production, the CO2 storage amount nearly stopped
increasing by the thirty-sixth year. Consequently, the CO2 storage efficiency is almost
equal to 1 at the beginning and then gradually decreases, as shown in Figure 6b.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 8 of 19

Figure 6. CO2 storage in the production process: (a) injection, production and storage amount of
Figure
CO2 , 6.and
CO(b)
2 storage in the production process: (a) injection, production and storage amount of
CO2 storage efficiency.
CO2, and (b) CO2 storage efficiency.

As shown in Figure 7, the overall effects of oil production on both CO2 flooding and
CO2-WAG are higher than those of water flooding. CO2 flooding has higher oil production
in the early stages of development before the thirty-sixth year because CO2 is more mobile
Appl. Sci. 2022, 12, 10958 8 of 18

As shown in Figure 5, at first, the oil production rate goes up quickly because there
is enough energy in the reservoir. As the reservoir energy gradually depletes, the oil
production curve begins to level off gradually in the fifth year. Then, the reservoir energy
is replenished by the nineth year after the CO2 -WAG started, and the oil production rate
begins to increase rapidly. The oil production rate begins to decrease slowly from the
twentieth year to the end of production.
At the beginning of the CO2 injection, the CO2 storage amount is almost equal to the
CO2 injection amount from the nineth year to the fifteenth year with little CO2 produced
(Figure 6a). CO2 production begins in the fifteenth year and has a relatively stable rate.
From the sixteenth year, the growth rate of CO2 storage starts to be lower than that of CO2
injection.
Figure 6. CO With the increase
2 storage in CO2 production,
in the production process: (a) the CO2 storage
injection, production amount nearlyamount
and storage stopped of
increasing
CO2, and (b) byCO the thirty-sixth
2 storage year. Consequently, the CO2 storage efficiency is almost equal
efficiency.
to 1 at the beginning and then gradually decreases, as shown in Figure 6b.
As shown
As shown in in Figure
Figure 7, 7, the
the overall
overall effects
effects ofof oil
oil production
production on onboth
bothCO CO22 flooding
flooding andand
CO22-WAG
CO -WAG are are higher
higher thanthan those
those of of water
water flooding.
flooding. CO CO22 flooding has higher oil productionproduction
in the
in the early
early stages
stages of of development
development before before the
the thirty-sixth
thirty-sixthyear
yearbecause
becauseCO CO22 is is more
more mobile
mobile
than
than water. CO CO2 2flooding
floodingproduces
produces oiloil
at aathigher
a higher raterate
thanthanbothboth
water water
floodingflooding
and COand2-
CO
WAG 2 -WAG
at at
this this
stage.stage.
Due Due
to to
the the heterogeneity
heterogeneity of of the
the reservoir,
reservoir, the
the CO
CO 22 injected
injected into the
the
formation
formationtendstendstotoform forma adominant
dominant channel.
channel. So,So,
thethe
COCO2 flooding
2 floodingmethod
method starts to
starts produce
to pro-
more
duce oil in aoil
more slow
in awayslow around the twenty-second
way around year of its
the twenty-second lateofstage
year of development.
its late stage of develop- The
CO
ment. -WAG method effectively combines the advantages of
2 The CO2-WAG method effectively combines the advantages of water flooding water flooding and CO 2 flooding
and
and
CO2 maintains
flooding and a high oil production
maintains a high oilrate, especially rate,
production after especially
the thirty-sixth
after theyear,thirty-sixth
although
the oilalthough
year, production therate is relativelyrate
oil production lower than that lower
is relatively of CO2thanflooding
that ofbefore the thirty-sixth
CO2 flooding before
year. The overall
the thirty-sixth oil production
year. The overall of oil CO 2 -WAG isofbetter
production CO2-WAGthan that of water
is better than flooding and
that of water
CO flooding.
flooding and CO2 flooding.
2

Figure 7. Cumulative
Figure 7. Cumulative oil
oil production
production of
of CO
CO22-WAG,
-WAG, CO
CO22 flooding
flooding and
and water
water flooding.
flooding.

For further analysis of the difference between the CO2 -WAG method and the CO2
flooding method, this work compares the gas–oil ratio of the produced flow. As shown in
Figure 8, the gas production rate starts to increase from the twelfth year (4 years after CO2
injection) in the CO2 flooding method and rises sharply from the nineteenth year onwards.
This can be explained by the fact that the injected CO2 forms a dominant channel, and thus
the injected CO2 returns to the surface directly from the dominant channel as an output gas.
 gas.
In the CO2-WAG method, the gas production rate is basically unchanged in the early
stage of production and increases at a low rate in the late stage of production. Thus, the
gas production volume is much smaller than that of CO2 flooding. It can be speculated
that the injected water in the CO2-WAG method effectively hinders the breakthrough of
Appl. Sci. 2022, 12, 10958 CO2 and the formation of a CO2-dominant channel. Therefore, CO2-WAG can effectively 9 of 18
slow down the rate of CO2 gas extraction compared to CO2 flooding.

Figure 8.
Figure Gas oil
8. Gas oil ratio
ratio of
of the
the production
production in
in the
the WAG
WAG and
and CO
CO22 flooding.
flooding.

In the CO2 -WAG

Currently, method, and
the treatment the gas production
separation of rate is basically
output gas to unchanged
recover methane in the early
and
stage of production and increases at a low rate in the late stage of production. Thus, the gas
reinjected carbon dioxide are priorities for many oilfields. Gas treatment stations in oil-
production volume is much smaller than that of CO2 flooding. It can be speculated that the
fields have a certain upper limit of gas that can be processed per day. In the CO2 flooding
injected water in the CO2 -WAG method effectively hinders the breakthrough of CO2 and
method, the gas output rate may be too high, especially in the final stages of development.
the formation of a CO2 -dominant channel. Therefore, CO2 -WAG can effectively slow down
It may lead to a risk that a large amount of methane gas and carbon dioxide cannot be
the rate of CO2 gas extraction compared to CO2 flooding.
captured and recovered in time, causing a great loss of economic benefits and more emis-
Currently, the treatment and separation of output gas to recover methane and rein-
sions of greenhouse gases. In contrast, according to the preceding analysis, the CO2-WAG
jected carbon dioxide are priorities for many oilfields. Gas treatment stations in oilfields
method can solve the aforementioned issues by efficiently extracting crude oil and reduc-
have a certain upper limit of gas that can be processed per day. In the CO2 flooding method,
ing the gas production rate, thereby allowing the gas treatment station sufficient time to
the gas output rate may be too high, especially in the final stages of development. It may
capture and recover methane gas and reinject carbon dioxide.
lead to a risk that a large amount of methane gas and carbon dioxide cannot be captured
and recovered in time, causing a great loss of economic benefits and more emissions of
3.2. Analysis of Influence Factors on CO2-WAG
greenhouse gases. In contrast, according to the preceding analysis, the CO2 -WAG method
To analyze
can solve the principal influence
the aforementioned factors of CO
issues by efficiently 2-WAG,
extracting this work
crude oil and designs
reducinga series of
the gas
scenarios of the CO -WAG development method by modifying the
production rate, thereby allowing the gas treatment station sufficient time to capture and
2 CO 2 -WAG period,
fluid injection
recover methane rate, and
gas andwater–gas ratio based
reinject carbon on the above base model. Among them, six
dioxide.
schemes of the period adjustment are set as 4 months, 6 months, 12 months, 24 months,
3.2.months,
48 Analysisandof Influence
96 months, Factors on CO2 -WAG
respectively. Six schemes of the fluid injection rate adjustment
are setToas: (1) water
analyze theinjection
principal rate of 65 m3factors
influence /day andof CO injectionthis
CO22 -WAG, rateworkof 12,000
designsm3/day; (2)
a series
water injection
of scenarios rateCO
of the of 79
2 m
-WAG 3 /day and CO
development 2 injection
method rate
by of 16,000
modifying m 3
the/day;
CO (3)
2 water
-WAG injec-
period,
tion
fluidrate of 90 m
injection 3/day and CO2 injection rate of 20,000 m3/day; (4) water injection rate of
rate, and water–gas ratio based on the above base model. Among them, six
schemes of the period adjustment are set as 4 months, 6 months, 12 months, 24 months,
48 months, and 96 months, respectively. Six schemes of the fluid injection rate adjustment
are set as: (1) water injection rate of 65 m3 /day and CO2 injection rate of 12,000 m3 /day;
(2) water injection rate of 79 m3 /day and CO2 injection rate of 16,000 m3 /day; (3) water
injection rate of 90 m3 /day and CO2 injection rate of 20,000 m3 /day; (4) water injection
rate of 100 m3 /day and CO2 injection rate of 23,000 m3 /day; (5) water injection rate of
110 m3 /day and CO2 injection rate of 27,000 m3 /day; (6) water injection rate 120 m3 /day
and CO2 injection rate of 31,000 m3 /day. 6 schemes of the water–gas ratio adjustment
are set as 0.33, 0.5, 1, 1.5, 2, and 3, respectively. Therefore, this work constructs a total of
6 × 6 × 6 = 216 scenarios. As the water injection rate varies simultaneously with the CO2
injection rate, this work selects the CO2 injection rate as one feature, and the period and
water–gas ratio as the other two features.
Appl. Sci. 2022, 12, 10958 10 of 18

The cumulative oil production, CO2 storage amount, and CO2 storage efficiency of
the 216 scenarios are simulated by CMG, and the results are detailed in Supporting Infor-
mation. To elucidate more clearly the influence of CO2 -WAG parameters on production,
the cumulative oil production, CO2 storage amount, and CO2 storage efficiency of differ-
ent cycle schemes (#85, #97, #103), different fluid injection rate schemes (#49, #121, #193),
and different water–gas ratio schemes (#85, #87, #89) are compared by the single variable
method, and the results are shown in Figures 9–11.
As shown in Figure 9, the influence of the cycle on oil production is relatively small.
When the CO2 -WAG cycle is relatively shorter, the cumulative oil production is higher.
When the cycle is 8 years, the cumulative oil production starts to decline significantly,
because in this case, the frequency of alternating water and carbon dioxide is low, similar
to a small period of water flooding and gas flooding. The influence of the cycle on CO2
storage is relatively large, both in terms of CO2 storage amount and CO2 storage efficiency.
Throughout the production process, CO2 storage amount and CO2 storage efficiency
fluctuate due to the alternate injection of water and CO2 . At the end of the process, the CO2
storage amount and CO2 storage efficiency decrease with the cycle.
As shown in Figure 10, the injection rate has a greater impact on oil production. As
the injection rate increases, cumulative oil production also increases, but the CO2 storage
efficiency decreases. This is because more CO2 will return to the surface from production
wells as output gas when the CO2 injection rate increases. Therefore, when optimizing the
CO2 -WAG extraction scheme in the oilfield, the injection rate cannot be increased arbitrarily.
The processing capacity of gas treatment stations in the oilfield needs to be considered.
When CO2 output is too fast, some CO2 will not be recovered and treated in time. CO2
will escape into the atmosphere, which may cause environmental issues and aggravate
the greenhouse effect. It also causes waste of CO2 gas resources and economic loss to the
oilfield. At the same time, an over-high injection rate will instantly increase the bottom
hole pressure of the injection well. When it exceeds the fracture pressure of the formation,
it will crush the formation, causing damage to the formation on the one hand. On the other
hand, it may cause CO2 to escape and pollute other formations.
The model in this study has a low reservoir pressure due to a period of depleted
extraction in the early stage. In addition, the fluid injection rate is low, which means it
cannot restore the reservoir pressure to the initial pressure. The simulated reservoir is a
low-pressure reservoir. For this type of reservoir, as shown in Figure 11, the higher the
water–gas ratio is, the better the oil production will be. In addition, when the injection rate
and the reservoir pressure are high, a lower water–gas ratio has a higher oil recovery. A
higher water–gas ratio results in less CO2 being buried because the proportion of injected
CO2 is smaller, but the storage efficiency is higher. A lower water–gas ratio allows more
CO2 to be buried because the proportion of CO2 in the injected fluid is larger. Moreover,
more CO2 will be produced from the production well, leading to lower storage efficiency.
Appl. Sci. 2022, 12, x FOR PEER REVIEW 11 of 19
Appl. Sci. 2022, 12, 10958 11 of 18

Figure 9. Effect of WAG cycle on (a) cumulative oil production, (b) CO2 storage amount, and (c) CO2
storage efficiency.
 Appl. Sci. 2022, 12, x FOR PEER REVIEW 12 of 19

Appl. Sci. 2022, 12, 10958 Figure 9. Effect of WAG cycle on (a) cumulative oil production, (b) CO2 storage amount, and 12
(c)of
CO182
storage efficiency.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 13 of 19

Figure 10. Effect of fluid injection rate on (a) cumulative oil production, (b) CO2 storage amount, and
(c) CO10.
Figure 2 storage
Effect efficiency.
of fluid injection rate on (a) cumulative oil production, (b) CO2 storage amount,
and (c) CO2 storage efficiency.
Appl. Sci. 2022, 12, 10958 13 of 18
Figure 10. Effect of fluid injection rate on (a) cumulative oil production, (b) CO2 storage amount,
and (c) CO2 storage efficiency.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 14 of 19

Figure 11. Effect of water–gas ratio on (a) cumulative oil production, (b) CO2 storage amount, and
(c) CO211.
Figure storage
Effectefficiency.
of water–gas ratio on (a) cumulative oil production, (b) CO2 storage amount, and
(c) CO2 storage efficiency.

3.3. Analysis of Machine Learning Results

To compare the CMG numerical simulation results with the machine learning regres-
 −11.93%, and 8.72%, respectively, the relative deviation for all other scenarios ranges from
−6.31% to 4.55%. The average absolute relative deviation for predicting CO2 storage effi-
ciency is 2.24%.
According to the curve between the predicted value and the true value, and their
relative deviation, the random forest regression algorithm can accurately predict the CO2-
Appl. Sci. 2022, 12, 10958 14 of 18
WAG production. Therefore, the cumulative oil production, CO2 storage amount, and CO2
storage efficiency can be accurately estimated by this method for the cases of various pe-
riods, fluid injection rates, and water–gas ratios.
3.3. Analysis
Moreover, of Machine
compared Learning
to the Results
simulation method, the machine learning-assisted pre-
dictionTomethod
comparecan thesaveCMG thenumerical simulation
time of adjusting theresults
CMG withmodel theand
machine
running learning regres-
data files. It
sion about
takes prediction results,
30 s to this of
run a case work buildsmodel
the CMG a database
and 108 ofmin
216 for
CMG numerical
all of simulation
the 216 cases in this
models,
study. Theofmachine
which 70% are used
learning as theemployed
method training set and work
in this 30% as the spends
only test set.10The numerical
s predicting
simulation
all results ofwhich
of the scenarios, the test set are
shows employed
a big as the
advantage in true values for verifying
computational efficiencythe accuracy
compared
of the machine learning prediction. The result plots for the three
to the CMG simulation method. If the grid of the CMG model is refined, the timesaving labels of cumulative
oil production,
advantage of theCO 2 storage
machine amount,
learning and CO
method 2 storage
will be much efficiency are shownthe
huger. Therefore, in Figure
machine 12.
The predicted curve made by the random forest regression algorithm
learning-assisted method can greatly improve the prediction efficiency of CO 2-WAG and is very close to the
truebe
will value curve
suitable forininjection
terms of cumulative
parameter oil production, CO2 storage amount, and CO2
optimization.
storage efficiency.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 16 of 19

Figure 12. Cont.

Appl. Sci. 2022, 12, 10958 15 of 18

Figure 12. Comparison of the predicted values by random forest algorithm with the true values in (a)
Figure 12. Comparison
cumulative of the
oil production, (b)predicted values
CO2 storage by random
amount, and (c)forest
CO2 algorithm with the true values in
storage efficiency.
(a) cumulative oil production, (b) CO2 storage amount, and (c) CO2 storage efficiency.
The relative deviation of cumulative oil production, CO2 storage amount, and CO2
storage efficiency for each scheme are calculated and shown in Figure 13 for further
discussion. The relative deviation of cumulative oil production between prediction and
true values ranges from −4.76% to 2.51%, the maximum relative deviation is −4.76%, and
the average absolute relative deviation is 1.10%. The maximum relative deviation of CO2
storage amount between prediction and true values is −15.15%. Except for this, the relative
deviation for all other scenarios ranges from −10.67% to 5.23%. The average absolute
relative deviation for predicting CO2 storage amount is 3.04%. For the prediction of CO2
storage efficiency, except for three scenarios with large relative deviations of −13.35%,
−11.93%, and 8.72%, respectively, the relative deviation for all other scenarios ranges
Figure
from 12. Comparison
−6.31% of the
to 4.55%. Thepredicted
averagevalues by random
absolute relativeforest algorithm
deviation with the true
for predicting COvalues in
2 storage
(a) cumulative
efficiency oil production, (b) CO2 storage amount, and (c) CO2 storage efficiency.
is 2.24%.

Figure 13. Cont.

Appl. Sci. 2022, 12, x FOR PEER REVIEW 17 of 19
Appl. Sci. 2022, 12, 10958 16 of 18

Figure 13. Relative deviation between the predicted value and true value in (a) cumulative oil
Figure 13. Relative
production, (b) CO2deviation betweenand
storage amount, the(c)
predicted valueefficiency.
CO2 storage and true value in (a) cumulative oil pro-
duction, (b) CO2 storage amount, and (c) CO2 storage efficiency.
According to the curve between the predicted value and the true value, and their
4. Conclusions
relative deviation, the random forest regression algorithm can accurately predict the CO2 -
WAGFor production. Therefore,
the application of CO2the-WAG cumulative oil production, COlow-permeability
used in non-homogeneous, 2 storage amount, and
reser-
CO storage efficiency can be accurately estimated by this method
voirs and the machine learning-assisted prediction method of CO2-WAG, some conclu-
2 for the cases of various
periods,
sions canfluid injection
be drawn fromrates,
this and
work water–gas ratios.
as follows:
(1) Compared to water flooding and continuousthe
Moreover, compared to the simulation method, COmachine learning-assisted predic-
2 flooding, CO2-WAG can effec-
tion method can save the time of adjusting the CMG model and
tively improve oil recovery. In addition, compared to CO2 flooding, CO2-WAG running data files.
can It takes
reduce
about
the CO302 production
s to run a case of the
rate, CMG
which model and to
is conducive 108the
min for all of
storage of the
CO216 cases in this study.
2 for the reduction of
The machine learning
greenhouse gas emissions. method employed in this work only spends 10 s predicting all of the
scenarios,
(2) The CO2-WAG cycle time has a slight influence on oil production. Both CO 2CMG
which shows a big advantage in computational efficiency compared to the stor-
simulation method. If the grid of the CMG model is refined, the timesaving advantage of
age amount and CO2 storage efficiency decrease with the cycle. On the premise that the
the machine learning method will be much huger. Therefore, the machine learning-assisted
reservoir formation is not fractured, the oil production increases but CO 2 storage effi-
method can greatly improve the prediction efficiency of CO2 -WAG and will be suitable for
ciency decreases with the fluid injection rate. For low-pressure reservoirs, the oil produc-
injection parameter optimization.
tion increases with the water–gas ratio, but CO2 storage efficiency decreases with the wa-
ter–gas ratio.
(3) The random forest regression algorithm in machine learning has better fitting ac-
curacy in predicting the results of CO2-WAG development. Therefore, it can be used to
Appl. Sci. 2022, 12, 10958 17 of 18

4. Conclusions
For the application of CO2 -WAG used in non-homogeneous, low-permeability reser-
voirs and the machine learning-assisted prediction method of CO2 -WAG, some conclusions
can be drawn from this work as follows:
(1) Compared to water flooding and continuous CO2 flooding, CO2 -WAG can effec-
tively improve oil recovery. In addition, compared to CO2 flooding, CO2 -WAG can reduce
the CO2 production rate, which is conducive to the storage of CO2 for the reduction of
greenhouse gas emissions.
(2) The CO2 -WAG cycle time has a slight influence on oil production. Both CO2
storage amount and CO2 storage efficiency decrease with the cycle. On the premise that the
reservoir formation is not fractured, the oil production increases but CO2 storage efficiency
decreases with the fluid injection rate. For low-pressure reservoirs, the oil production
increases with the water–gas ratio, but CO2 storage efficiency decreases with the water–
gas ratio.
(3) The random forest regression algorithm in machine learning has better fitting
accuracy in predicting the results of CO2 -WAG development. Therefore, it can be used
to predict the oil production and CO2 storage results under different combinations of
injection parameters.
(4) Compared to numerical simulations, using machine learning algorithms to predict
results avoids the need to build models and run data files, which will save a lot of time for
subsequent parameter optimization.

Supplementary Materials: The following are available online at https://www.mdpi.com/article/10

.3390/app122110958/s1.
Author Contributions: Conceptualization, H.L.; Data curation, C.G.; Formal analysis, C.G.; Funding
acquisition, H.L., J.X. and S.L.; Investigation, C.G.; Methodology, H.L. and C.G.; Project admin-
istration, H.L. and S.L.; Software, C.G.; Supervision, S.L.; Writing—original draft, H.L. and C.G.;
Writing—review & editing, H.L., S.L., J.X. and G.I. All authors have read and agreed to the published
version of the manuscript.
Funding: This research was funded by the National Natural Science Foundation of China [52074337,
51906256, 52174052, 51904323], the Shandong Provincial Natural Science Foundation [ZR2021JQ18]
and Fundamental Research Funds for the Central Universities [21CX06033A].
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: All data is involved in the text.
Acknowledgments: This research is supported by the National Natural Science Foundation of China,
the Shandong Provincial Natural Science Foundation and Fundamental Research Funds for the
Central Universities, which are gratefully acknowledged.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Gao, J.; Lv, J. Problem analysis on low permeability reservoir water flooding. Inn. Mong. Petrochem. Ind. 2009, 48–51.
2. Li, S.; Tang, Y. Present situation and development trend of CO2 injection enhanced oil recovery technology. Oil Gas Reserv. Eval.
Dev. 2019, 9, 1–8.
3. Chen, H.; Liu, X. Prospects and key scientific issues of CO2 near-miscible flooding. Pet. Sci. Bull. 2020, 3, 392–401.
4. Liu, S.; Ren, B.; Agarwal, B. CO2 storage with enhanced gas recovery (CSEGR): A review of experimental and numerical studies.
Pet. Sci. 2022, 19, 594–607. [CrossRef]
5. Wei, Q.; Hou, J. Laboratory study of CO2 channeling characteristics in ultra-low-permeability oil reservoirs. Pet. Sci. Bull. 2019, 2,
145–153.
6. Chen, Z.; Su, Y. Characteristics and mechanisms of supercritical CO2 flooding under different factors in low-permeability
reservoirs. Pet. Sci. 2022, 19, 1174–1184. [CrossRef]
7. Tang, R.; Wang, H. Effect of water and gas alternate injection on CO2 flooding. Fault Block Oil Gas Field 2016, 23, 358–362.
Appl. Sci. 2022, 12, 10958 18 of 18

8. Sanchez, N.L. Management of water alternating gas (WAG) injection projects. In Proceedings of the Latin American and Caribbean
Petroleum Engineering Conference, Caracas, Venezuela, 21–23 April 1999.
9. Christensen, J.R.; Stenby, E.H. Review of WAG field experience. SPE Reserv. Eval. Eng. 2001, 4, 97–106. [CrossRef]
10. Rochedo, P.R.R.; Costa, I.V.L. Carbon capture potential and costs in Brazil. J. Clean. Prod. 2016, 131, 280–295. [CrossRef]
11. Godoi, J.M.A.; dos Santos Matai, P.H.L. Enhanced oil recovery with carbon dioxide geosequestration: First steps at Pre-salt in
Brazil. J. Pet. Explor. Prod. 2021, 11, 1429–1441. [CrossRef]
12. Kengessova, A. Prediction efficiency of immiscible Water Alternating Gas Performance by LSSVM-PSO algorithms. Master’s
Thesis, University of Stavanger, Stavanger, Norway, 2020.
13. Mohagheghian, E.; James, L.A. Optimization of hydrocarbon water alternating gas in the Norne field: Application of evolutionary
algorithms. Fuel 2018, 223, 86–98. [CrossRef]
14. Bender, S.; Yilmaz, M. Full-Field Simulation and Optimization Study of Mature IWAG Injection in a Heavy Oil Carbonate
Reservoir. In Proceedings of the SPE Improved Oil Recovery Symposium, Tulsa, OK, USA, 12–16 April 2014.
15. Johns, R.T.; Bermudez, L. WAG optimization for gas floods above the MME. In Proceedings of the SPE Annual Technical
Conference and Exhibition, Denver, CO, USA, 5–8 October 2003.
16. Chen, S.; Li, H. Optimal parametric design for water-alternating-gas (WAG) process in a CO2 -miscible flooding reservoir. J. Can.
Pet. Technol. 2010, 49, 75–82. [CrossRef]
17. Mousavi Mirkalaei, S.M.; Hosseini, S.J. Investigation of Different I-WAG Schemes Toward Optimization of Displacement Efficiency.
In Proceedings of the SPE Enhanced Oil Recovery Conference, Kuala Lumpur, Malaysia, 19–21 July 2011.
18. Ghaderi, S.M.; Clarkson, C.R. Evaluation of recovery performance of miscible displacement and WAG process in tight oil
formations. In Proceedings of the SPE/EAGE European Unconventional Resources Conference and Exhibition, Vienna, Austria,
20–22 March 2012.
19. You, J.; Ampomah, W. Assessment of Enhanced Oil Recovery and CO2 Storage Capacity Using Machine learning and Optimization
Framework. In Proceedings of the SPE Europec featured at 81st EAGE Conference and Exhibition, London, England, UK, 3–6
June 2019.
20. Zhou, D.; Yan, M. Optimization of a mature CO2 flood—From continuous injection to WAG. In Proceedings of the SPE Improved
Oil Recovery Symposium, Tulsa, OK, USA, 14–18 April 2012.
21. Rodrigues, H.; Mackay, E.; Arnold, D.; Silva, D. Optimization of CO2 -WAG and Calcite Scale Management in Pre-Salt Carbonate
Reservoirs. In Proceedings of the Offshore Technology Conference Brasil, Rio de Janeiro, Brazil, 29–31 October 2019.
22. Min, C.; Dai, B. A Review of the Application Progress of Machine Learning in Oil and Gas Industry. J. Southwest Pet. Univ. 2020,
42, 1–15.
23. He, Y.; Song, Z. Application of machine learning in hydraulic fracturing. J. China Univ. Pet. 2021, 45, 127–135.
24. Bilgesu, H.I.; Al-Rashidi, A.F. An unconventional approach for drill-bit selection. In Proceedings of the SPE Middle East Oil Show,
Manama, Bahrain, 17–20 March 2001.
25. Leite Cristofaro, R.A.; Longhin, G.A. Artificial Intelligence Strategy Minimizes Lost Circulation Non-Productive Time in Brazilian
Deep Water Pre-Salt. In Proceedings of the OTC Brasil, Rio de Janeiro, Brazil, 24–26 October 2017.
26. Wang, W.; Shi, M. Joint optimization method of well location and injection-production parameters based on machine learning. J.
Shenzhen Univ. 2022, 39, 126–133.
27. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
28. Feng, M.; Yan, W. Predicting Total Organic Carbon Content by Random Forest Regression Algorithm. Bull. Mineral. Petrol.
Geochem. 2018, 37, 475–481.
29. Zhen, Y.; Hao, M. Research of medium and long term precipitation forecasting model based on random forest. Water Resour.
Power 2015, 33, 6–10.
30. Wang, P.; Lu, B. Water demand prediction model based on random forests model and its application. Water Resour. Prot. 2014, 30,
34–37.
31. Lv, H.; Feng, Q. A review of random forests algorithm. J. Hebei Acad. Sci. 2019, 36, 37–41.
32. Du, X.; Feng, J. PM2.5 concentration prediction model based on random forest regression analysis. Telecommun. Sci. 2017, 33,
66–75.
33. Zhang, Y. The Study on Reservoir Description and Remaining Oil Distribution on the 4th–6th Sandstone Beds of the Second Member in the
Shahejie Formation of Tuo 28 Block in Shengtuo Oilfield; China University of Petroleum: Qingdao, China, 2017.
34. Miao, C. Reservoir Description on the 1st–3rd Sandstone Beds of the Second Member in the Shahejie Formation of the Tuo 28 Block in
Shengtuo Oilfield; China University of Petroleum: Qingdao, China, 2016.
35. Cutler, A.; Cutler, D.R.; Stevens, J.R. Random Forests. In Ensemble Machine Learning; Zhang, C., Ma, Y., Eds.; Springer: Boston,
MA, USA, 2012.
36. Fang, K.; Wu, J. A Review of Technologies on Random Forests. J. Stat. Inf. 2011, 26, 32–38.
37. Hillebrand, E.; Lukas, M. Bagging weak predictors. Int. J. Forecast. 2021, 37, 237–254. [CrossRef]
38. Criminisi, A.; Shotton, J. (Eds.) Decision Forests for Computer Vision and Medical Image Analysis; Springer Science & Business Media:
Berlin/Heidelberg, Germany, 2013.