You are on page 1of 9

SPE-188100-MS

Probabilistic Performance Forecasting in History Matching of Low


Permeability Reservoirs
Xuejiao Zhang, Yongchao Xue, Guanyang Ding, China University of Petroleum, Beijing

Copyright 2017, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Kingdom of Saudi Arabia Annual Technical Symposium and Exhibition held in Dammam, Saudi Arabia, 24–27
April 2017.

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
Since conventional history matching aimed at fractured horizontal wells in low permeability oil reservoirs
is affected by a number of factors, such as permeability, fracture half-length, conductivity and so on, there
often exists ambiguity in production matching. That is to say, when showing the same curves and results,
we cannot make a definite decision to judge which parameter displays the matching results. In this paper,
Markov Chain Monte Carlo (MCMC) and AM algorithm are presented to improve history matching and
then to obtain more accurate probabilistic production forecasting using actual decline production data.
First, the AM algorithm, having an advantage of updating parameters simultaneously and constituting a
proposal distribution at each new iteration according to the covariance matrix of the previous iterations
over the Metropolis-Hasting (MH) algorithm, is employed to gather field production decline data.
Furthermore, MCMC method is utilized to develop a Markov Chain, a stochastic process with a series of
various parameters and later value usually only related with the most-recent value. Finally, based on this,
the history matching is improved and further probabilistic production prediction P10, P50 and P90 are
achieved.
The results indicate that compared with MH algorithm, the AM algorithm can get a greater
acceptance ratio. The Markov Chain for production decline data parameters shows a satisfying mixing.
The probabilistic cumulative oil production of P10, P50, and P90 is established for target oilfield in this
paper. The curves of production rate versus time and cumulative oil production versus time show that the
well-established Markov Chain can successfully match the production decline data and then perfectly
predict probabilistic production. The novel point in this paper is that a much more effective AM algorithm
substituting for the MH algorithm is adopted to form the Markov Chain to improve history matching. The
results manifest that MCMC method has the ability to enlarge the reliability of production forecasts, which
has a significant influence on reservoir understanding and management.
Introduction
Low permeability oil resources distribute in almost all oil-producing countries, whose reserves are
abundant, consequently, they have been effectively developed and employed in the world. In China, low
permeability reservoirs are rich in oil reserves, and the proved reserves are nearly 6 billion tons. However,
due to the complex structure, poor permeability, heterogeneity of low permeability reservoirs, nearly 50%
of the proved low permeability reserves are difficult to use because of low oil recovery. At present,
fractured horizontal well technology is the significant means to develop low-permeability reservoirs. By
means of forming a number of cracks in the horizontal wells, it is possible to enhance production capacity
2 SPE-188100-MS

and improve development effect of low-permeability oilfields due to expanding the drainage area of single
wells. Hence, it is of importance to forecast production of low permeability reservoirs through history
matching to guide the oilfields operations. Usually, the numerical simulation is conducted by setting up a
geological model, then inputting relative reservoirs parameters to obtain the static distribution of porosity
and permeability (Cheng et al.2010), at last matching the production to forecast future production. Yet,
the production uncertainty still cannot avoid in production matching and prediction since there exist
complicated fracture structures and flow patterns (Clarkson and Qanbari 2014; Yu et al. 2016). Moreover,
conventional history matching is controlled by many factors, such as fracture half-length, permeability,
conductivity and so on, it is hard to judge which parameter displays the matching results if getting the
same curves and results, as a result, conventional history matching is no longer proper to predict
production accurately and objectively. Even though the numerical simulation is advanced and improved
to consider these uncertainty factors, it is a little a waste of time to operate the models to match the
production of quite a few wells (Wei. 2016). Instead, many authors (Giger. 1985; Wan.1999; Guo. 2006;
Yuan and Zhou. 2010; Hao. 2013) do the research about the analytical models equations to predict
production aimed at low permeability oil reservoirs, but many complex mechanisms and factors are
simplified to make equations solving easier, which could also lead to some uncertainties when evaluating
performance (Wei. 2016). In order to make a quantitative uncertainty in the process of prediction, the
combination of Markov Chain Monte Carlo (MCMC) and AM algorithm in history matching is a robust
and high-effective method to quantify uncertainty (Cheng et al.2010) to obtain more reliable probabilistic
production forecasting.
MCMC is a Monte Carlo method which is simulated by computer under the framework of Bayesian
theory, which has a frequent and wide application in quantifying production uncertainty (Vink et al. 2015).
Oliver et al. (1997) introduced MCMC into single-phase flow condition to obtain the posterior probability
distribution. Christen and Fox (2005) made an improvement aimed at MCMC method to adopt
approximation model to ignore the declined transitions. Efendiev et al. (2008) introduced sensitivity
evaluation into MCMC method to estimate parameters. Gong et al. (2014) proposed a Bayesian
probabilistic method together with MCMC. Arps (1945) employed decline-curve model analysis. They
made a conclusion that their method and theory could use a narrower P10–P90 interval to quantify
production uncertainty. Combining with updating proxy models to quantify a reliable and satisfying
production probabilistic uncertainty, Goodwin (2015) improved MCMC method by using a creative
Hamiltonian MCMC (Hoffman and Gelman2014) technology, which obtains convergence in a faster rate
compared with MH algorithm. Nowadays, Monte Carlo simulation method has transcended many
deterministic methods to become a superior technology to predict production (Williamson et al. 2006). In
the paper, Markov Chain Monte Carlo (MCMC) sampling is used to improve the history matching of low
permeability oil production data and get the posterior distribution of decline-curve parameters.
Methodology
Markov Chain Monte Carlo (MCMC)
The Markov Chain Monte Carlo method is an important stochastic method associated with statistical
physics, which is widely used in Bayesian inference and machine learning. In essence, the MCMC method
is the use of the Markov Chain Monte Carlo integral. The MCMC method based on Bayesian inference
principle is mainly applied to generate posterior distribution samples, calculate the edge distribution and
posterior distribution matrix.
Assume: k-dimensional random vector U  (U1, ,Uk ) has a joint distribution  (U1 , ,Uk ) , and U k is
a model parameter or a missing observation value in the Bayesian statistical application. The mathematical
expectation for the function h U  is as follows:
E h U    h(u) (u)du (1)
In practice, h U  is often complex and difficult to calculate, so Monte Carlo integral is used to get
approximation results, that is:
SPE-188100-MS 3

1 n
E  h U     h(U (t ) )
n t 1
(2)

When U1 , ,Uk are independent with each other, by law of large numbers, the larger the sample size
n is, the higher the degree of its approximation is. However, in many complicated models, the
independence assumptions of U1 , ,Uk cannot be made simply, consequently, the Markov Chain Monte
Carlo simulation method is needed to deal with the problem.
Markov Chain is a model owing the simplest process and the most application. For a random sequence
t , t  T  , if the probability of value of t 1 state is only associated with the state  t , we call it the Markov
Chain which is shown next:
p(t 1  A 0 , ,t )  p(t 1  A t ) (3)
Where A represents state space of  t ; T usually represents the time scale.
The basic idea of MCMC method is to establish a Markov Chain to simulate unknown variables U k and to
obtain the needed posterior distribution when the chain is under the condition of steady distribution.
Random point U k is derived from the distribution  (U ) , and different MCMC methods are obtained from
different sampling methods.
Monte Carlo method is a numerical method with a unique style, both in terms of steps of the method,
accuracy and convergence of the results. The advantages of Monte Carlo method and the difference
compared with general numerical methods can be summarized as the following three aspects:
(1) The program structure of it is simple, which calculates the integral achieved by a large number of
simple and repetitive sampling.
(2) The probability of convergence and convergence rate are independent of the dimension of
problems.
(3) Monte Carlo method is highly adaptable, which is nearly not limited by the impact of problems
conditions.
Monte Carlo forecast can be regarded as applying statistical sampling process to supply approximate
solutions of complicated conditions and issues, which has been put into use to a number of cases in our
petroleum area, such as engineering design (Sawaryn et al. 2002), besides production evaluation and
prediction.
AM algorithm
Aiming at the problem that the search rate of Metropolis-Hastings is slow, Haario et al (2001) presented
an adaptive Metropolis algorithm. Compared with the traditional Metropolis-Hastings algorithm, adaptive
Metropolis (AM) no longer needs to determine the proposal distribution of parameters in advance, but
sample the posterior distribution through the covariance matrix of parameters. The covariance matrix of
posterior parameter adjusts itself adaptively after each iteration. The proposal distribution of the step i
parameters is defined as the multivariate and normal distribution of mean  i and covariance Ci .
The calculation of covariance matrix is in the following formula:
 C0 i  i0
Ci   (4)
 sd Cov 0 ,1 , ,i 1   sd  I d i  i0

Where C0 is the initial covariance, and under the condition of initial sampling times i  i0 , covariance is
given a fixed value C0 in order to eliminate the instability of the algorithm sample at the initial stage;  is a
smaller constant to ensure that Ci is not a singular matrix; sd is a scale factor, depending on spatial
dimension d of parameters, which is suggested: sd   2.4 / d ( Gelman A C, et al. 1996); I d is D-
2

dimensional unit matrix.


For the time of (i+1)th iteration, the covariance formula can be deduced from the following equation:
4 SPE-188100-MS

i 1
Ci  d ii 1i T1  (i  1)ii T  iiT   I d 
s
Ci 1  (5)
i i
Where i 1 is the mean of the previous time of (i-1)th iteration parameters;  i is the mean of the previous
time of i iteration parameter.
The one key content of the adaptive Metropolis (AM) is to define the proposal distribution of
parameters of the i th step, and the another key issue is to define acceptance probability to judge and
determine whether to accept the newly generated subsamples. The acceptance probability can be defined
as:
 p( q  * ) p( * ) 
  min 1,  (6)
p( q i ) p(i ) 
 
Where q refers to the decline production data of low permeability oil reservoirs in this paper;  * is
given based on the existing  deriving from a proposal distribution; p( ) represents prior distribution of
parameter  ; p( q  ) shows the likelihood function between parameter  and production data q ; In
addition, in comparison with p( q  ) , p( q) is the posterior distribution of parameter  .
The biggest advantage of the AM algorithm is that the proposal distribution can be automatically
updated with calculation process and no longer needs to be specified. In comparison with the traditional
MH algorithm, the parameters can be updated at the same time, and the group update is no longer
necessary, which reduces computation greatly.
Workflow
In order to adopt the methodology above to improve history matching of low permeability oil reservoirs
production, the steps and procedures are designed as follows. And Figure 1 displays the workflow in detail
clearly.
Step 1: Set the initial proposal distribution  0 and the covariance matrix C0 ;
Step 2: Begin iteration step as i  1 ;
Step 3: Form new proposal distribution  * ;
 p( q  * ) p( * ) 
Step 4: Calculate acceptance probability   min 1, ;
 p( q i ) p(i ) 
Step 5: Examine by Equation (4) and (5) to judge whether or not to accept the newly proposal distribution.
If Yes, go forward directly to step 6; If no, go back to step 3;
Step 6: Update the currently existing state to the newly proposed distribution;
Step 7: Continue to the (i+1)th iteration;
Step 8: Examine if i has satisfied the criteria. If Yes, then stop the procedures; If no, employ Equation (4)
to regain Ci , and after that, go back to step 3.

Fig. 1—Workflow of methodology combination in the paper.


SPE-188100-MS 5

Application to a Low Permeability Oil Reservoir


To forecast probabilistic performance in history matching of low permeability oil reservoirs with the
method of combination of Markov Chain Monte Carlo (MCMC) and AM algorithm, a field case is
conducted in Changqing low permeability oilfield of China. The target oilfield has an average permeability
between 10  103 μm2 and 80  103 μm2 , which has been put into development and production since the year
of 1998 with a total of 221 oil wells up to now. The target oilfield owns geological reserves of 850.69 104 t
and the degree of recovery of 9.51%. (Dong. 2014).
Field production data gathering
To gather the target oilfield actual decline production data which is typical in quality and large in quantity,
some principles should be followed:
There should gather production data plenty enough to express a complete range of production
capacity and results so as to eliminate the influence of low sampling size.
There should contain the production data coming from the wells which are related to the subject of
the prediction in the target oilfield to guarantee that the collected data can contribute to reasonable
forecasting results.
History matching using MCMC and AM algorithm
By collecting and organizing the target oilfield actual decline production data, it is determined that the
data conforms to the Arps hyperbolic decline characteristics, whose equation is as follows:
1

q  qi (1  Di bt ) b
(7)
Where b is the hyperbolic decline exponent, whose value generally is [0, 1], and in reservoir engineering,
it owns oilfield definition; Di represents initial decreasing rate. In the paper, b is restricted to [0, 1] and Di
is a positive number or Di  0 .
Based on above, b and Di are two decline-curve parameters. As for their prior distribution, we can
set them according to existing understanding about them (Lucy et al. 2004). Here, we can also assume that
the two parameters have an identical prior distribution. As for their posterior distribution, we can get them
using the MCMC sampling and history production data. In this target field case, MCMC and AM
algorithm are combined to perform history matching. After that, with confidence intervals of P10-P90,
probabilistic production forecasts can be obtained.

Fig. 2—History matching combining MCMC and AM algorithm.

However, the iteration numbers should be considered reasonably to converge to the final posterior
distribution. Here, 6000 iterations are selected to make a prediction to the two parameters’ posterior
distribution. Generally, i0 is set as 100 and covariance matrix Ci can only be adjusted and updated at
every 100th step. In our practice, the initial value of the two decline-curve parameters b and Di are
respectively defined as b  0.623 , Di  0.205 . When 6000 history matching is over together with AM
6 SPE-188100-MS

algorithm, we acknowledge 2216 cases as satisfying history matches as is shown in Fig. 2, which has an
acceptance ratio of 36.93%.

0.7

b 0.6

0.5
0 1000 2000 3000 4000 5000 6000
Iteration Number
Fig. 3—Markov Chain of parameters b with 6000 iterations.

0.6

0.4
Di

0.2

0
0 1000 2000 3000 4000 5000 6000
Iteration Number
Fig. 4—Markov Chain of parameters Di with 6000 iterations.

As is shown is Fig. 3 and Fig. 4, together with AM algorithm, we respectively display the Markov
Chain condition in every iteration of the two decline-curve parameters b and Di with a total of 6000
iterations. From the two figures, we can judge easily that b changes in the region [0.52, 0.74], while Di
moves between 0.13 and 0.58. We can also conclude deeply that the two parameters’ Markov Chains are
not limited in some specified area of their spaces, but distributed evenly and stochastically, which can
imply that the Markov Chains shows a satisfactory mixing outcome. Subsequently, the two parameters’
posterior distribution are forecasted based on their change interval, which is shown in Fig. 5 (a) and (b)
respectively. We can note from the figure, compared with parameter b, parameter Di has a more stable
and well-distributed distribution. By MCMC sampling, the transformation from the prior distribution to
the posterior distribution of the two decline-curve parameters b and Di is finished so that the aim is
obtained that the production forecast risk decreases to the least, which make a guarantee for history
matching.
120
100

90
75
Density
Density

60
50

30 25

0 0
1 3 5 7 9 11 13 15 17 0.6
19 21 23 25 27 29 31 33 35 0.7
37 39 41 43 0.75
45 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
0.5 0.1 0.3 0.5 0.6
b Di
(a) (b)
SPE-188100-MS 7

Fig. 5—Posterior distributions of the two parameters b and Di .

Probabilistic production prediction


From the history matching above with the MCMC simulation and the AM algorithm, the two parameters’
posterior distribution are achieved successfully. Next, combing the forecasted quantifying posterior
distribution, production prediction for 24 years were conducted. Here, the confidence interval of P10, P50
and P90 is adopted to forecast the target oilfield production, which especially shows the plots of P10, P50
and P90 in Fig. 6. According to the figure, the green line is about the actual production data, the purple
area is the confidence interval of the predicted production and the three blue lines are P10, P50 and P90
from the bottom up.

Fig. 6—Cumulative oil production forecasting for 24 years with level of P10, P50 and P90.

In addition, cumulative oil production of different confidence intervals at 24 years based on Fig. 6
is illustrated in Fig. 7 and the corresponding cumulative oil production at 24 years for P10, P50, P90 is
marked clearly, which is 3002, 3519, 4025 STB in turn.
1
(P90,4025)
Confidence interval

0.75

(P50,3519)
0.5

0.25
(P10,3002)

0
2000 2500 3000 3500 4000 4500
Cumulative Oil Production/STB
Fig. 7—Cumulative oil production of different confidence intervals at 24 years.

Similarly, the production prediction outcomes of other wells in the target oilfield also confirm this
probabilistic prediction method along with MCMC sampling and AM algorithm has a universal
application value, which means this approach can conduct production data analysis, history matching and
forecast in other low permeability oil reservoirs, maybe even in shale gas reservoirs.
Conclusions
1. Markov Chain Monte Carlo (MCMC) sampling, the AM algorithm and confidence level of P10, P50
and P90 are coupled together to forecast production in history matching of low permeability oil
reservoirs.
8 SPE-188100-MS

2. The AM algorithm can update parameters simultaneously and no longer need to determine the
proposal distribution of parameters in advance and covariance matrix of posterior parameter adjusts
itself adaptively after each iteration.
3. The basic idea of MCMC method is to establish a Markov Chain to simulate unknown variables and
to obtain the needed posterior distribution when the chain is under the condition of steady distribution.
4. The Markov Chain for two decline-curve parameters b and Di shows a satisfying mixing, and b
changes in the region [0.52, 0.74], while Di moves between 0.13 and 0.58.
5. The probabilistic production prediction is quantified by the confidence levels, that is to say, the
corresponding cumulative oil production for P10, P50, P90 is forecasted and obtained respectively.
Nomenclature
A = state space of  t
b = the hyperbolic decline exponent
C0 = initial covariance
Ci = covariance matrix
Di = the initial decreasing rate
i0 = an index for the initial period of non-adaptation
I d = d-dimensional unit matrix
n = the sample size
p( ) = the prior distribution of parameter 
p( q  ) = the likelihood function between parameter  and production data q
p( q) = the posterior distribution of parameter 
q = the decline production data of the target low permeability oil reservoir
sd = a scale factor
T = the time scale
U k = a model parameter
U  (U1 , ,Uk ) = k-dimensional random vector
 = acceptance probability
 = a smaller constant
 (U1 , ,Uk ) =a joint distribution
 = a model parameter
 * =the value based on the existing  deriving from a proposal distribution
i 1 = parameter’ mean of the previous (i-1) iteration
 i = parameter’ mean of the previous i iteration
Acknowledgements
This paper has been supported by National Natural Science Foundation of China (No.41672132). The
authors thank all the senior fellow apprentices and peers for their supports and encouragement to the paper
from the bottom of their hearts. At the same time, the authors would also like to give special and sincere
thanks to Wang Kaiyi for his enthusiastic help and instructions about English writing and grammar.
References
Approach for Addressing the Limitations of Existing Empirical and Analytical Methods. Presented at the SPE Annual Technical Conference
and Exhibition, Amsterdam, 27–29 October. SPE-170767-MS. http://dx.doi.org/10.2118/170767-MS.

Arps, J. J. 1945. Analysis of Decline Curves. J Pet Technol 160 (1): 228–247. SPE-945228-G. http://dx.doi.org/10.2118/945228-G.

Clarkson, C. R. and Qanbari, F. 2014. A Semi-Analytical Forecasting Method for Unconventional Gas and Light Oil Wells: A Hybrid.

Christen JA, Fox C. 2005. Markov Chain Monte Carlo using an approximation. J Comput Graph Stat 14(4):795–810.
SPE-188100-MS 9

Cheng, Y., Wang, Y., McVay, D. A. et al. 2010. Practical Application of a Probabilistic Approach to Estimate Reserves Using Production
Decline Data. SPE Econ & Mgmt 2 (1): 19–31. SPE-95974-PA. http://doi.org/10.2118/95974-PA.

Dong Linyan. 2014. Studies on Indoor Chemical Oil Displacement System of Changqing Low Permeability Reservoir [D]. China University
of Petroleum(East China).

Efendiev Y, Datta-Gupta A, Ma X, Mallick B (2008) Modified Markov Chain Monte Carlo method for dynamic data integration using
streamline approach. Math Geosci 40:213–232.

Giger, F. M. 1985. Horizontal Wells Production Techniques in Heterogeneous Reservoirs. Presented at the SPE Middle East Oil Technical
Conference and Exhibition, 11-14 March, Bahrain. SPE-13710-MS. https://doi.org/10.2118/13710-MS.

Gelman A C, Roberts G O, Gilks W R. 1996. Efficient Metropolis jumping rules. In J. M.Bernardo, J. O. Bergen A. F. David and A.
F.M.Smith (eds), Bayesian Statistics V, 599-608. Oxford: Oxford University Press.

Gong, X., Gonzalez, R., McVay, D. A. et al. 2014. Bayesian Probabilistic Decline-Curve Analysis Reliably Quantifies Uncertainty in
ShaleWell-Production Forecasts. SPE J. 19 (6): 1047–1057. SPE-147588-PA. http://dx.doi.org/10.2118/147588-PA.

Goodwin, N. 2015. Bridging the Gap Between Deterministic and Probabilistic Uncertainty Quantification Using Advanced Proxy Based
Methods. Presented at the SPE Reservoir Simulation Symposium, Houston, 23–25 February. SPE-173301-MS.
http://dx.doi.org/10.2118/173301-MS.

Guo, J., Zeng, F., Zhao, J.. et al. 2006. A New Model to Predict Fractured Horizontal Well Production. Presented at the PETSOC Canadian
International Petroleum Conference, Calgary, Alberta, 13-15 June. PETSOC-2006-076. https://doi.org/10.2118/2006-076.

Haario, H., Saksman, E., and Tamminen, J. 2001. An Adaptive Metropolis Algorithm. Bernoulli 7: 223–242.

Hoffman, M. D. and Gelman, A. 2014. The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo. Journal of
Machine Learning Research 15 (1):1593–1623. http://jmlr.org/papers/volume15/hoffman14a/hoffman14a.pdf.

Hao, M., Hu, Y., Liu, X.. et al. 2013. Predicting and Optimising the Productivity of Multiple Transverse Fractured Horizontal Wells in Ultra-
Low Permeability Reservoirs. Presented at the International Petroleum Technology Conference, Beijing, China, 26-28 March. IPTC-
16891-Abstract .https://doi.org/10.2523/IPTC-16891-Abstract.

Williamson, H. S., Sawaryn, S. J., & Morrison, J. W. 2006. Monte Carlo Techniques Applied to Well Forecasting: Some Pitfalls. SPE Drilling
& Completion, 21(3), 44-58, September.SPE-89984-PA.https://doi.org/10.2118/89984-PA.

Lucy, M., Nott, D., and Sharma, A. 2004. A Comparative Study of Markov Chain Monte Carlo Methods for Conceptual Rainfall-Runoff
Modeling. Water Resources Research 40 (2): 1–11. http://dx.doi.org/10.1029/2003WR002378.

Oliver D, Cunha L, Reynolds A. 1997. Markov Chain Monte Carlo methods for conditioning a permeability field to pressure data. Math Geol
29:61–91. http://dx.doi.org/10.1007/BF02769620.

Sawaryn, S.J., Grames, K.N., and Whelehan, O.P. 2002. The Analysis and Prediction of Electric-Submersible-Pump Failures in the Milne
Point Field, Alaska. SPE Production & Facilities, 17 (1): 53–61, February. SPE-74685-PA. https://doi.org/10.2118/74685-PA.

Vink, J. C., Gao, G., and Chen, C. 2015. Bayesian Style History Matching: Another Way to Under-Estimate Forecast Uncertainty? Presented
at the SPE Annual Technical Conference and Exhibition, Houston, 28–30 September. SPE-175121-MS.
http://dx.doi.org/10.2118/175121-MS.

Wadsley, A. W. 2005. Markov Chain Monte Carlo Methods for Reserves Estimation. Presented at the International Petroleum Technology
Conference, Doha, Qatar, 21-23 November. IPTC-10065-MS. https://doi.org/10.2523/IPTC-10065-MS.

Wan, J., & Aziz, K. 1999. Multiple Hydraulic Fractures in Horizontal Wells. Presented at the SPE Western Regional Meeting, Anchorage,
Alaska, 26-27 May. SPE-54627-MS. https://doi.org/10.2118/54627-MS.

Yuan, H., & Zhou, D. 2010. A New Model for Predicting Inflow Performance of Fractured Horizontal Wells. Presented at the SPE Western
Regional Meeting, Anaheim, California, USA, 27-29 May. SPE-133610-MS. https://doi.org/10.2118/133610-MS.

Yu, W., Sepehrnoori, K., and Patzek, T. W. 2016. Modeling Gas Adsorption in Marcellus Shale with Langmuir and BET Isotherms. SPE J.
21 (2): 589–600. SPE-170801-PA. http://dx.doi.org/10.2118/170801-PA.

Zhang, Y., & Srinivasan, S. 2003. Markov Chain Monte Carlo for Reservoir Uncertainty Assessment. Presented at the Canadian International
Petroleum Conference, Calgary, Alberta, 10-12 June. PETSOC-2003-084. https://doi.org/10.2118/2003-084.

You might also like