IPTC-21821-MS

A Data-Driven Approach to Evaluate Fracturing Practice in Tight Sandstone

in Changqing Field

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Tao Wu, CNPC Chuanqing Drilling Engineering Co.LTD; Hanzhi Fang, Yangtze University; Hu Sun, CNPC
Chuanqing Drilling Engineering Co.LTD; Feifei Zhang, Xi Wang, Yidi Wang, and Siyang Li, Yangtze University

Copyright 2021, International Petroleum Technology Conference

This paper was prepared for presentation at the International Petroleum Technology Conference held virtually on 23 March - 1 April 2021. The official proceedings
were published online on 16 March 2021.

This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submitted by the author(s).
Contents of the paper, as presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The
material, as presented, does not necessarily reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at
IPTC are subject to publication review by Sponsor Society Committees of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial
purposes without the written consent of the International Petroleum Technology Conference 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 where and by whom the paper was presented.
Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435.

Abstract
Unconventional reservoirs such as shale and tight sandstones that with ultra-low permeability, are becoming
increasingly significant in global energy structures (Pejman T, et al., 2017). For these reservoirs, successful
hydraulic fracturing is the key to extract the hydrocarbon resources efficiently and economically. However,
the intrinsic mechanisms of fracturing growth in the tight formations are still unclear. In practice, fracturing
design mainly depends on hypothetical models and previous experience, which leads to difficulties in
evaluating the performance of the fracturing jobs. Therefore, an improved method to optimize parameters
for fracturing is necessary and beneficial to the industry.
In this paper, a data-driven approach is used to evaluate the factors that dominate the production rate
from tight sandstone formation in Changqing Field which is the largest oil field in China. In the model, the
input parameters are classified into two categories: controllable parameters (e.g. stage numbers, fracturing
fluid volume) and uncontrollable parameters (e.g. formation properties), and the output parameter is the
accumulated oil production of the wells. Data for more than 100 wells from different formations and zones
in Changqing Field are collected for this study. First, a stepwise data mining method is used to identify the
correlations between the target parameter and all the available input parameters. Then, a machine learning
model is developed to predict the well productivity for a given set of input parameters accurately.
The model is validated by using separate data-sets from the same field. An optimize algorithm is
combined with the data-driven model to maximize the cumulative oil production for wells by tuning the
controllable parameters, which provides the optimized fracturing design. By using the developed model,
low productivity wells are identified and new fracturing designs are recommended to improve the well
productivity.
This paper is useful for understanding the effects of designed fracturing parameters on well productivity in
Changqing Oilfield. Furthermore, it can be extended to other unconventional oil fields by training the model
with according data sets. The method helps operators to select more effective parameters for fracturing
design, and therefore reduce the operation costs for fracturing and improve the oil and gas production.
2 IPTC-21821-MS

Introduction
In China, tight oil accounts for about two-fifths of all the recoverable oil resources and is considered to be
one of the most realistic energy resources in the future. The Chang-7 formation in Longdong area is a typical
tight oil reservoir, which has the characteristics of tight lithology, small pore throats, and large percolating
resistance (Zhao Z.F, et al., 2019). For tight oil reservoirs, the "horizontal well drilling + volume fracturing"
technology has become a key method for hydrocarbon resources extraction. In the past, fracturing operation
design in oilfield mainly relied on establishing a fracture propagation model and verifying it by either lab

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

experiments or numerical simulations. This process involves a number of impractical assumptions, and the
results are often questionable. The data mining method can be used to discover the relationship between
fracturing parameters and the final production rate from the historical data, and provide theoretical basis for
the optimization of fracturing design by establishing accurate prediction models.
In recent years, a variety of data mining methods have been used for fracturing design and operation.
Nejad, Amir (Nejad, A. M, et al., 2015) used neural network models trained by multiple databases to
comprehensively assess the parameters related to completion and fractures, and established a model for
oil and gas production, which was verified by eight wells from eagle ford field. Pejman Tahmasebi et al.
(Pejman T, et al., 2017) used data mining and machine learning techniques to identify the sweet spots for
shale reservoirs. Mohaghegh et al. (Mohaghegh, S. D, et al., 2017) applied a data-driven technology in
the design of secondary fracturing schemes for Niobrara, Eagle Ford and Marcellus wells and achieved a
completion optimization method by analyzing the generated typical curves. Aniemena et al. (Aniemena,
C, et al., 2017) proposed a process for selecting candidate wells for fracturing based on advanced data-
driven analysis method. Results show that this method can dramatically improve the efficiency of drilling
operations and provide useful information for fracturing deign during the re-fracturing stage. McVey, D.S
(McVey, D. S, et al., 1994) designed, trained and applied an artificial neural network to evaluate the outcome
of hydraulic fracturing design.
This paper investigates the effects of the dominate factors (which include reservoir parameters and
fracturing operational factors) on well productivity. An optimized Support Vector Machine (SVM) method is
used to establish the production prediction model for tight oil reservoirs in the Longdong area of Changqing
Oilfield. The outcomes of this work can provide guidance for tight oil fracturing design.

Sensitivity Analysis
First, this paper determines input parameters of the production prediction model by analyzing the correlation
between the production after fracturing and the collected formation and construction parameters. Then,
after comparing the regression effects of three data-driven models (including elastic network regression,
decision tree regression and support vector regression) applicable to small-scale data sets, the support
vector regression algorithm was selected as the production prediction model for fracturing evaluation of
the research block.
In this work, data from 136 wells in the research block of the Longdong area are collected. The wells are
distributed in 6 formations (mainly distributed in chang-7 formation in Figure 1) and have been in production
for the maximum of seven years. In order to more comprehensively and scientifically analyze the factors
that typically affect the production after fracturing in the area, a qualitative analysis of the relationship
between the collected parameters and the production has been conducted, which is the critical information
to determine the number of input parameters for the production prediction model.
IPTC-21821-MS 3

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 1—chang-7 formation reservoir profile

The various parameters closely related to the production of 136 wells in the area have been sorted
out, including permeability, oil saturation, pumping rate, the length of horizontal section, horizontal
section number, the distance between fractured stages, clustering space, fracturing fluid, and the proppant
properties.
This paper initially chose to analyze the relationship between cumulative oil production and parameters.
For horizontal wells, the horizontal lateral length is an important factor affecting productivity. As the length
of horizontal section increases, the productivity of horizontal wells increases rapidly. The relationship
between the length of horizontal section and the oil production in the first year is shown in Figure 2, and the
influence of the length of horizontal section on the overall cumulative oil production is shown in Figure 3.

Figure 2—Effect of the length of horizontal section on cumulative oil production of the first year
4 IPTC-21821-MS

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 3—Effect of the length of horizontal section on the overall cumulative oil production

It can be seen from Figures 2 and Figures 3 that the horizontal section lengths of old wells are generally
greater than 1500 m, and the horizontal section lengths of new wells are generally less than 1000 m. And the
length of horizontal section has an obvious linear relationship with the cumulative oil production. Therefore,
it is concluded that there is still room for increasing the length of horizontal section, because it has not
reached the length that will limit the production rate due to excessive downhole flow. So, the following
analysis needs to use the cumulative oil production of unit horizontal section length (tons/m) to remove the
influence of horizontal section length and obtain the influence of other parameters on production accurately.
The following analysis focuses on analyzing the relationship between the cumulative oil production of
unit horizontal section length and proppant, fracturing fluid, the distance between fractured stages, cluster
spacing and other parameters.
According to the 136 oil wells in the investigated block, five types of proppants are mainly used, including
20/40 mesh quartz sand (A), 20/40-40/70 mesh quartz sand (B), 20/40 mesh quartz Sand with 40/70 mesh
low density ceramsite (C), 20/40 mesh quartz sand with 40/70 mesh and 20/40 mesh consolidated sand (D),
and quartz sand with 20/40 mesh low density ceramsite (E).
Figure 5 is an enlarged diagram of the cumulative oil production of unit horizontal section length and
production days of the three types of proppants in Figure 4. It can be found that for the new wells in the
past three years, the performance of two (A, B) proppants are not much different, and the performance of
the third (C) proppants is significantly worse than the two proppants.
IPTC-21821-MS 5

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 4—Cumulative oil production of unit horizontal section length for five types of proppant

Figure 5—Cumulative oil production of unit horizontal section length under three types of proppant

The effect of fracturing fluids on the cumulative oil production rate is shown in Figure 6. Four fracturing
fluid systems include "EM30S" (A), "EM30" (B), "EM30" slickwater + guanidine gum (C) and guanidine
gum (D). It can be seen from Figure 6 that the performance of type C and D fracturing fluids are not as good
as the other two types, and the performance of type A fracturing fluids is relatively better than the other three.
6 IPTC-21821-MS

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 6—Cumulative oil production of unit horizontal section length with four fracturing fluid

As shown in Figure 7, the distance between fractured stages is negatively correlated with the cumulative
oil production of unit horizontal section length. As shown in Figure 8 and Figure 9, from 140 m to 40
m, as the distance between fractured stages decreases, the cumulative oil production of unit horizontal
section length increases significantly. However, further reducing the distance between fractured stages has
no obvious effect on the production once the 40 m. It can be indicated that optimal stage distance is about
40 m.

Figure 7—The effect of the distance between fractured stages on the production of the first year
IPTC-21821-MS 7

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 8—The effect of the distance between fractured stages on overall cumulative oil production

Figure 9—Optimal distance between fractured stage

As shown in Figure 10, the cluster spacing is negatively correlated to the cumulative oil production of
unit horizontal section length in the first year. It can be seen that from Figure 11 and Figure 12 that from 50
m to 3 m, as the cluster spacing decreases, the cumulative oil production of unit horizontal section length
increases significantly. However, further reducing the cluster spacing has no obvious effect on production
once it is less than 10 m. It can be concluded that the cluster spacing of the optimal output is about 10 m.
8 IPTC-21821-MS

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 10—The effect of cluster spacing on cumulative oil production of unit horizontal section length in the first year

Figure 11—The effect of cluster spacing on the overall cumulative oil production of unit horizontal section length
IPTC-21821-MS 9

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 12—The effect of cluster spacing on the overall cumulative oil production of unit horizontal section length

From Figure 13, it can be found that porosity is strongly correlated with the overall cumulative oil
production. From Figure 14, it can be found that permeability has an obvious positive correlation with the
overall cumulative oil production of unit horizontal section length. From Figure 15, it can be found that
the oil saturation has a more obvious positive correlation with the overall cumulative oil production of unit
horizontal section length.

Figure 13—The effect of porosity on the overall cumulative oil production of unit horizontal section length
10 IPTC-21821-MS

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 14—The effect of permeability on the overall cumulative oil production of unit horizontal section length

Figure 15—The effect of oil saturation on the overall cumulative oil production of unit horizontal section length

In summary,10 parameters have been analyzed, including the length of horizontal section, proppant
specification, fracturing fluid, the distance between fractured stages, cluster spacing, pumping rate, oil
saturation, porosity, permeability and production month. These 10 parameters have been chosen to be the
input parameters of the production prediction model, and the analysis of the parameter influence will be
used as guidance for optimizing production. Information on proppant specifications and fracturing fluids
are textual parameters. In order to take these two types of factors into data-driven model as input parameters,
these text information are artificially numbered and digitized.
IPTC-21821-MS 11

Model Selection for Data Mining Method

Data mining is a data analysis method which uses machine learning methods to collect, clean, analyze, and
model data samples. The purpose of the approach is to generalize and conclude the regularity contained
in the data, and then establish qualitative or quantitative mathematic model to predict the properties of the
unknown sample. This paper attempts to use data mining algorithms to conduct sensitivity analysis of the
parameters which affect the oil productivity of the fractured wells. The ultimate goal is to obtain a qualitative
analysis and quantitative prediction model which can be consistent with the oilfield results, and use it as

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

a reference for future fracturing design.
One of the major problems for applying data-driven method in oil and gas industry is the shortage of the
labeled data. As the available useful data is limited, the use of data-driven methods based on large-scale
data sets cannot achieve ideal results. Support vector regression, elastic net regression and decision tree
regression, which require less data volume, are more applicable. A brief introduction of these three data-
mining methods is given below.
Elastic net regression is a combination of "Lasso regression" and "Ridge regression". Ridge regression
adds a regular term (namely the square deviation factor) on the basis of standard linear regression, which
makes the optimization function min||Xω − y||2 + z||ω||2. Lasso is very similar to "Ridge regression", and it
also adds a bias term to the optimization function to reduce the effect of collinearity. The difference is that
the absolute value deviation is used as the regular term in "Lasso regression", and its optimization function
is min||Xω − y||2 + z||ω||, which also makes the generalization ability of the model stronger and can help
solve the model problem of "overfitting".
The basic algorithm of decision tree regression is the greedy algorithm, that is, the top-down recursive
divide and conquer method. The algorithm is generally divided into two stages: "tree generation and
pruning". The latter is to remove unnecessary branches from the tree and solve "over-learning" problem
(Tan F.Q, et al., 2010). The decision tree in practical applications may be very complicated, but the meaning
of each path from the root node to the leaf node can be understood, which is a significant advantage over
other methods. Therefore, decision trees are widely used in knowledge discovery systems.
The basic algorithm of support vector (SVR) regression is to find the inner relationship between a
bunch of data. By fitting the data at high latitudes, the algorithm can get a formula, and when a new input
value is given, a new output value can be obtained. The biggest difference between SVR regression and
traditional regression methods is: traditional regression methods require that the prediction is correct only
if the returned f(x) is exactly equal to y, while SVR regression believes that when f(x) deviates from y, if the
degree is within a certain range, the prediction is considered correct. The specific is to set a threshold value
α and calculate the loss of data points with |f(x)− y| > α. For the SVR model, the most important parameter is
the type of kernel, which generally includes linear kernel, polynomial kernel, hyperbolic tangent kernel and
gaussian radial basis function (rbf), etc. Due to the non-linear problem, this paper chose rbf as the kernel
function. The parameters that mainly affect rbf include the penalty coefficient C and the kernel parameter γ.
In order to qualitatively and quantitatively evaluate the regression effects of the three models on fracturing
optimization problems, cross-validation (Schuettler J, et al., 2018) is chosen to evaluate the advantages and
disadvantages of the regression results of the three methods on fracturing optimization. The remaining data
are used as input to predict production after fracturing. The regression error value of three models are shown
in Table 1. The training data and the comparison of the test data of these three models are shown in Figure

16-18. The definition of R2 in Figure 16-18 is where y is the actual production of

the oil wells and f is the model forecast production, is the average value of y. When the value of R2 is
approximately close to 1, the effect of model regression is better. And according to the definition, R2 may
be less than 0, so R2 does not represent the square of R value.
 Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022
IPTC-21821-MS

Figure 17—elastic network regression

Figure 16—support vector regression
12
IPTC-21821-MS 13

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 18—decision tree regression

Table 1—Comparison of the regression results of the three models

Actual Forecast Cumulative Oil production/(105tons)

Cumulative Oil
Number Support Vector The relative Elastic Net The relative Decision Tree The relative
production /
(105tons) Regression error/% Regression error/% Regression error/%

Well No 1 0.355677 0.3090 13.11 0.5701 60.29 0.4090 15.00

Well No 2 0.061303 0.1256 104.84 0.0745 21.48 0.0562 8.33

Well No 3 0.232276 0.2094 9.86 0.3093 33.17 0.2248 3.24

Well No 4 0.289892 0.3024 4.32 0.3453 19.12 0.2458 15.21

Well No 5 0.477347 0.4909 2.84 0.5194 8.81 0.5622 17.78

Well No 6 0.907003 0.5824 35.78 0.5460 39.80 0.7669 15.45

Well No 7 0.325996 0.3200 1.85 0.3834 17.61 0.2458 24.60

Well No 8 0.788072 0.7978 1.23 0.8023 1.81 0.7779 1.29

Well No 9 0.167116 0.1533 8.25 0.3753 124.56 0.2323 38.98

Well No 10 0.744901 0.6428 13.70 0.7262 2.50 0.6983 6.25

In addition to qualitatively observing the advantages and disadvantages of the model's output results, the
variance, mean absolute error, mean square deviation and coefficient of determination of regression model
have been calculated to evaluate the effect of the three models, as shown in Table 2. It is shown that the
effect of support vector regression and decision tree regression is better than the effect of elastic network
regression.

Table 2—Evaluation scores of model effect indicators

Explained variance score Mean absolute error Mean squared error r2 score

Support Vector Regression 0.94173 0.04160 0.01232 0.8927

Elastic Net Regression 0.69432 0.10617 0.02344 0.6819

Decision Tree Regression 0.85626 0.05371 0.00446 0.9094

14 IPTC-21821-MS

In conclusion, it can be found that support vector machine algorithm has obvious advantages for small
samples and complex nonlinear data. Therefore, SVM is chosen to conduct an exploratory analysis of the
fracturing effect of tight oil reservoirs as the optimization model.
As shown in Figure 19, a "Grid Search" method (Mohaghegh, S. D, et al., 2016) is applied to determine
the two-parameter of the support vector machine algorithm, which are the optimal penalty factor C and the
kernel parameter γ. In the two-dimensional parameter matrix composed of C and γ, the effects of each pair of

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

parameters are successfully tested. First, the ranges of the optimal penalty factor C and the kernel parameter
γ are given as (2−10,210.2) and (2−10,28.2), and the calculation step length is both 0.2. The regression variance
score of the model is shown in Figure 19. It can be seen from the figure that the optimal combination of C
and γ is C = 0.5,γ = 0.09, which corresponding to the maximum score

Figure 19—parameter optimization results

Case Study
After the establishment of the tight sandstone production prediction model and the analysis of production
influencing factors in the Longdong area of Changqing oilfield, two oil wells were selected to demonstrate
the effects of fracturing parameter optimization on production. Meanwhile, in order to describe the
productivity changes more directly, the production prediction model was used to predict the initial decline
in production and the initial decline rate of oil wells, thereby a decline curve was obtained.

Well#1
The formation and completion parameters for well#1 are shown in Table 3 and the actual production rate
of the oil well is about 5.83 kilotons/year. The prediction from the data-driven model is 6.05 kilotons/year.
Comparing the actual and forecasted productivity, it can be found that the predicted value of the model is
close to the actual value, and the prediction error is within 4% (Figure 20). After optimizing the fracturing
parameters by using an optimization algorithm, the distance between fractured stages is changed from 63.7
m to 40 m, and the cluster spacing is changed from 30 m to 10 m. With the optimized parameters, the
predicted production rate is 6.47 kilotons/year. The optimized result is about 11%higher than the original
predicted production rate.
IPTC-21821-MS 15

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 20—Fracturing optimization results of "Well#1"

Table 3—Input factors of "Well#1"

affecting factors the historical value the optimized value

the length of horizontal section 1531.44 1531.44

proppant 8 8

fracturing fluid 4 4

the distance between fractured stages 63.7 40

cluster spacing 20 10

pumping rate 3 3

oil saturation 57.84 57.84

porosity 9.4 9.4

permeability 0.14 0.14

production month 78 78

Well#2
The formation and completion parameters of Well#2 are listed in Table 4. The actual production rate of
the oil well is about 5.15 kilotons/year. The prediction from the data-driven model is 6.17 kilotons/year.
Comparing the height of the histogram (Figure 21), it can be found that the prediction from the data-
driven model is close to the actual production rate, and the prediction error is within 10%. After optimizing
the fracturing parameters by using an optimization algorithm, the space of horizontal section is changed
from102.7 m to 40 m, the cluster spacing is changed from 20 m to 10 m, and the length of the horizontal
16 IPTC-21821-MS

section is increased to 1400 m within the controllable cost range. With the optimized parameters, the
predicted production rate is 10.55 kilotons/year. The optimized result is about 25% higher than the original
predicted production rate.

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

Figure 21—Production prediction results of "Well#2"

Table 4—Input factors of "Well#2"

affecting factors the historical value the optimized value

the length of horizontal section 854 1400

proppant 7 7

fracturing fluid 4 4

the distance between fractured stages 102.7 40

cluster spacing 20 10

pumping rate 3 3

oil saturation 53.3 53.3

porosity 8.56 8.56

permeability 0.88 0.88

production month 77 77

Conclusion
This paper proposes a data mining method for the production prediction of the tight sandstone block in
Longdong area of Changqing Oilfield. The purpose is to evaluate the fracturing efficiency which has
multiple influence factors that interacting with each other, and obtain more reliable data from the perspective
IPTC-21821-MS 17

of "data-driven". The high-performance production prediction model provides certain guidance for on-site
fracturing operation in Longdong area of Changqing Oilfield.
1. By comparing the three modeling methods of elastic network regression?decision tree regression
and support vector regression. Among them, support vector machine algorithm can not only carry
out qualitative analysis modeling, but also can carry out multiple quantitative regression modeling.
Especially for small samples, non-linear complex data SVM has its unique advantages. Therefore, the
support vector regression method was selected for production prediction. To improve the accuracy of

Downloaded from http://onepetro.org/IPTCONF/proceedings-pdf/21IPTC/7-21IPTC/D071S024R002/2426929/iptc-21821-ms.pdf/1 by OMV E&P GmbH user on 18 December 2022

model prediction, "cross-validation" was used for the model parameters optimization. Among them,
the multiple modeling methods provided in data mining greatly expand the methods of fracturing
evaluation.
2. The optimized completion parameters for Longdong area are: 40 m stage distance, 10 m cluster
spacing, and2.5 m3/min pumping rate.

References
1. Aniemena, C., Kohshour, I. O. (2017, July 24). A Fast Semi-Analytical Method for Refracturing
Candidate Selection and Performance Estimation of Shale Wells.[C] Unconventional Resources
Technology Conference. doi:10.15530/URTEC-2017-2693452
2. Liu C. B. (2017) Single well productivity dynamic evaluation and characteristic analysis of
continental tight oil reservoirs [D]. China University of Petroleum (in Chinese).
3. McVey, D. S., Shahab M. S., Aminian, K. (1994, January 1). Identification of Parameters
Influencing the Response of Gas Storage Wells to Hydraulic Fracturing With the Aid of a Neural
Network [J]. Society of Petroleum Engineers. doi:10.2118/29159-MS
4. Mohaghegh, S. D. (2016, August 1). Fact-Based Re-Frac Candidate Selection and Design in
Shale A Case Study in Application of Data Analytics. [C] Unconventional Resources Technology
Conference. doi:10.15530/URTEC-2016-2433427
5. Mohaghegh, S. D., Gaskari, R., & Maysami, M. (2017, January 24). Shale Analytics: Making
Production and Operational Decisions Based on Facts: A Case Study in Marcellus Shale[J].
Society of Petroleum Engineers. doi:10.2118/184822-MS
6. Nejad, A. M., Sheludko, S., Shelley, R. F., Hodgson, T., Mcfall, P. R. (2015, February 3). A Case
History: Evaluating Well Completions in Eagle Ford Shale Using a Data-Driven Approach [J].
Society of Petroleum Engineers. doi:10.2118/173336-MS.
7. Pejman T., Farzam J., Muhammad S. (2017) Data mining and machine learning for identifying
sweet spots in shale reservoirs[J]. Expert Systems With Applications, 88.
8. Schuettler, J., Mishra et al. (2018, August 1). A Data-Analytics Tutorial: Building Predictive
Models for Oil Production in an Unconventional Shale Reservoir. Society of Petroleum
Engineers. doi:10.2118/189969-PA
9. Tan F. Q., Li H. Q., Meng Z. X., Guo H. F., Li X. Y. (2010) Application of data mining
methods in petroleum exploration and development[J]. Petroleum Geophysical Prospecting (in
Chinese),45 (01): 85-91+164+172.
10. Zhao Z. F., Bai X. H., et al. (2019) Production prediction method of volume fracturing horizontal
wells based on fuzzy set theory[J]. Oil Drilling & Production Technology (in Chinese), 41 (04):
521–528.