SPE-192354-MS

Comparing 5-Different Artificial Intelligence Techniques to Predict Z-factor

Adel Salem, A. Elgibaly, M. Attia; Suez Univesity, Abdulazeez Abdulraheem; KFUPM

Copyright 2018, 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, 23–26
April 2018.

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
Gas compressibility factor plays an important role in reservoir engineering applications. A lot of
techniques have been proposed to predict Z-factor. Standing-Katz (S-K) Z-factor chart is the most
common and popular among them and is being used since 1941. Many correlations have been proposed
after S-K chart to regenerate and increase its range in an accurate manner. Some of these models are direct
models such as Papp Correlation, Shell Oil Company Correlation, and Beggs and Brill Correlation, others
are indirect correlations such as Hall-Yarborough and Dranchuk-Abu-Kassem Correlation.

In this study, five different artificial intelligence techniques are implemented to predict Z-factor. These
techniques are neural network, radial basis function network, fuzzy logic, functional network, and support
vector machine. To build and test these techniques, Standing-Katz charts data was used in which about
70% of the data was used for training and 30% for testing.

Results from this work show that artificial intelligence techniques can predict Z-factor with low error such
as Neural network, Radial basis function, Fuzzy logic, and Support vector machine. Neural network is
the best technique among others in predicting Z-factor.

This work will help in selecting the best artificial intelligence technique for predicting Z-factor.

Introduction
Compressibility factor is the ratio between actual volume to ideal volume. Its value indicates how much
the real gas deviates from the ideal gas behavior at a certain pressure and temperature. Z-factor values are
mainly used in reservoir engineering (Xiang 2005). Accurate calculation of Z-factor affects determination
of other properties for the case of gas condensate and gas reservoirs. The Standing-Katz charts have been
used as a standard to calculate Z-factor in petroleum engineering since 1942. These charts were developed
using the concept of pseudo-reduced properties, reduced temperature, Tr, and reduced pressure, Pr.

𝑇 𝑃
𝑇𝑟 = & 𝑃𝑟 =
𝑇𝑐 𝑃𝑐

A lot of research have been conducted after Standing-Katz charts trying to fit these charts and to extend
them. There are several correlations available to predict Z-factor [Standing and Katz 1942, Papay 1968,
SPE-192354-MS 2

Beggs and Brill 1979, Burnett 1979, Papp 1979, Hall and Yarborough 1973, Mahmoud 2013, Lateef 2013,
Elechi et. al 2015].

These correlations can be classified into direct correlations such as Shell Oil Company Correlation (Kumar
2004), Papp Correlation (Papp 1979), Beggs and Brill Correlation (Beggs and Brill 1979) and Iterative
correlations such as Dranchuk-Abu-Kassem Correlation (Dranchuk and Abou-Kassem 1975) and Hall-
Yarborough Correlation (Hall and Yarborough 1974). These correlations proved good estimation for Z-
factor in certain range while the error was high when different ranges of data were used.

In our study, artificial intelligence (AI) methods are used to predict compressibility factor (Z-factor) and
compare the results obtained from different AI methods such as radial basis function network (RBF),
artificial neural network (ANN), fuzzy logic (FL), functional network (FN), and support vector machine
(SVM).

In the 1940s, petroleum industry has seen the application of optimization methods, especially in
exploration and production. After that it was applied for different problems in petroleum engineering such
as asphaltene precipitation, well placement, wettability, history matching, drilling operations, etc., (Wang
2003). Fig. 1 shows a classification diagram representing different optimization techniques. Artificial
Intelligence (AI) techniques have seen applied in a lot of petroleum engineering problems. Artificial
intelligence is defined as the ability of intelligent agents to perform continuous learning in the
corresponding environment and perceiving certain activities (Jang et al. 1997).

Fig. 1— Optimization techniques classification (Mirzabozorg 2015; Mohagheghian 2016)

Artificial Neural Networks (ANNs) consist of an input and output layers and layer between them (hidden
layers). Each hidden layer consists of number of nodes. Every node has certain weight which is adjusted
during the learning process to produce the expected output (Demuth and Beale 2002).
SPE-192354-MS 3

Fuzzy Logic (FL) is a technique used for uncertain reasoning to mimic the aspect of human cognition as
introduced by Dr. Lotfi Zadeh in the 1960’s for quantifying uncertainty (Zadeh 1965). Fuzzy logic refers
to something not well defined. several methods have been proposed in the literature. Adaptive Neuro-
Fuzzy Inference System (ANFIS) method is used in this paper (Nauck et al. 1997).

Radial basis function (RBF) describes any real-valued function whose output value depends exclusively
on the distance of its input from some origin. The radial basis function (RBF) networks is a class of
feedforward neural networks. It is used widely in different fields such as pattern recognition and seismic
exploration (Lei et. al 2011).

Support Vector Machines (SVM) Support Vector Machine (SVM) classification method was introduced
by Vapnik in 1963(Jakkula 2006). Support vector machine (SVM) is a supervised machine learning
method capable of deciphering subtle patterns in noisy and complex datasets.

Functional networks (FN) The functional network is a generalization of the standard neural network that
replaces sigmoidal standard models with generalized functional types. The neuron functions associated
with each neuron in functional networks are learned from the available data (Oladele et. Al 2014).
Z-Factor model construction

Matlab software code was developed to construct the different AI Techniques, viz., neural network
(ANN), radial basis function network (RBF), fuzzy logic, functional network (FN), and support vector
machine (SVM).

The Standing-Katz charts data, Fig. 2, were used in constructing the different artificial intelligence models.
70% of the data is used for the training and the remaining 30% for the testing for all the AI techniques.
Reduced pressure (Pr) and reduced temperature (Tr) were used as inputs and Z-factor (Gas compressibility
factor) was used as the output.

Fig. 2— Standing and Kats Compressibility Factor Chart (Ahmed 2006)

SPE-192354-MS 4

Results and discussions:

ANN model
Different number of layers and neurons for each layer were used to reach the optimum ANN model which
can predict Z-factor with low error. Optimized ANN using newff (feed forward back propagation) model
resulted in an average absolute percentage error of 0.294 and 1.056 for training and testing, respectively.
This was achieved with two hidden layers having 5 neurons for each layer. Fig. 3 shows Standing-Katz
charts data with blue colour and the predicted one from the ANN model with red colour. As can be seen
from the figure, the predictions are satisfactory.

Fig. 3— Regression between ANN predicted data and Standing-Katz charts data

Fuzzy Logic (FL) model

Fig. 4 shows the average absolute percentage error for training and testing data with different cluster
radius using ANFIS model

Fig. 4— comparison between different cluster radius and average absolute percentage error for the training and the testing Z-factor
predicted data using ANFIS model
SPE-192354-MS 5

It can be seen from Fig. 4 that the predictions with a cluster radius of 0.3 gave best results, with an average
absolute percentage error of 0.84 and 3.29 for training and testing, respectively. Fig. 5 shows regression
between the Standing-Katz data and the predicted one from the FL model. It can be seen from Fig. 5 that
Fuzzy logic ANFIS predicted data is little away from the Standing-Katz charts data.

Fig. 5— Regression between FL ANFIS (0.3 cluster radius) predicted data and Standing-Katz charts data

RBF Model:
The RBF model based on generalized regression neural network (newgrnn) code yielded an average
absolute percentage error of 0.9 and 3.9 for training and testing, respectively. Fig. 6 shows regression
between the Standing-Katz data and the predicted one from the RBF model.

Fig. 6— Regression between RBF predicted data and Standing-Katz charts data
SPE-192354-MS 6

SVM Model:
SVM model was also used to predict Z-facor. This resulted in an average absolute percentage error of
0.73 and 4.7 for training and testing respectively. Fig. 7 shows regression between the Standing-Katz data
and the predicted one from the SVM model.

Fig. 7— Regression between SVM predicted data and Standing-Katz charts data

FN Model:
A number of models were explored using functional network technique. However, none of those methods
worked satisfactorily. Fig. 8 shows regression between the Standing-Katz data and the predicted one from
FN model.

Fig. 8— Regression between FN predicted data and Standing-Katz charts data

SPE-192354-MS 7

Conclusion
In this paper we compared the performance of different artificial intelligence techniques for predicting Z-
factor using Standing-Katz charts data. For each AI technique, different methods were tried by changing
regression parameters. The following are the conclusions that can be drawn from this study:

1) Artificial intelligence neural network can predict Z-factor with an average absolute percentage error
lower than other AI techniques.
2) Fuzzy logic (ANFIS), radial basis function (newgrnn), and support vector machine (gaussian) can
also predict Z-factor but with a relatively higher average percentage error.
3) Functional network was not able to predict Z-factor Standing-Katz charts data properly. The average
absolute percentage error was high compared with other AI techniques.

Acknowledgements
The authors of this paper would like to thank Petroleum Engineering Depattment Professors, Suez
University and King Fahd University of Petroleum and Minerals (KFUPM)

References
Ahmed, T. 2006. “Reservoir Engineering HandBook, 3rd Edition”. Gulf Professional Publishing.
Beggs, H. and Brill, J. 1978. Two-Phase Flow in Pipes. Tusla, OK, The University of Tusla.
Burnett, R. R. 1979.Calculator Gives Compressibility Factors. The Oil and Gas Journal, 70−74.
Demuth, H. and Beale, M. 2002. Neural Network Toolbox™ User's Guide
Dranchuk, P.M., Abou-Kassem, J.H., 1975. Calculation of z factors for natural gases, using equations of state. J. Can. Petrol. Technol. 14
(3): 34-36. https://doi.org/10.2118/75-03-03
Elechi, V. U., Ikiensikimama, S.S. and Azubuike, I.I. 2015.A Correlation for estimating Gas Compressibility Factor in the Niger Delta.
Festschrift for J. A. Ajienka, 137−148.
Hall K.R., Yarborough L.1974. How to Solve Equation of State for Z Factors. Oil & Gas J, 86−88.
Hall, K. R., and Yarborough L. J. 1973. A New Equation-of-State for z-factor Calculations. Oil and Gas Journal, 82−92.
Jakkula, V. 2006. Tutorial on Support Vector Machine (SVM). School of EECS, Washington State University.
Jang J-SR, Sun C-T, Mizutani E.1997. Neuro-fuzzy and Soft Computing: A Computational Approach To Learning And Machine Intelligence.
IEEE Trans, 42(10):1482–1484
Kumar, N., 2004. Compressibility Factor for Natural and Sour Reservoir Gases by Correlations and Cubic Equations of State. MS thesis.
Texas Tech University, Lubbock, Tex, USA.
Lateef, A. K. 2013. Explicit Half Range Fourier Series Expansion for z Factor. Presented at Nigeria Annual International Conference and
Exhibition held in Lagos, Nigeria. 5-7 August. SPE-167579-MS. https://doi.org/10.2118/167579-MS
Lei, L., Wei, X., Shifan, Z. and Zhonghong, W.2011. Reservoir Property Prediction Using the Dynamic Radial Basis Function Network.
Mahmoud, M. A. 2013.Development of A New Correlation of Gas Compressibility Factor For High Pressure Gas Reservoirs. Presented at
The North Africa Technical Conference and Exhibition, Cairo, Egypt, 14-16 September. SPE-175724-MS.
https://doi.org/10.2118/164587-MS
Mirzabozorg, A.2015. Incorporation of Engineering Knowledge In History Matching, Optimization, And Uncertainty Assessment
Frameworks with Application to The SAGD Process. University of Calgary, Calgary
Mohagheghian, E.2016. An Application of Evolutionary Algorithms for WAG Optimisation in the Norne Field, Memorial University of
Newfoundland
Nauck Detlef; Klawonn, Frank; Kruse, Rudolf. 1997. Foundations of Neuro-Fuzzy Systems. John Wiley & Sons.
Oladele Bello and Tesleem Asafa, Belgium. 2014. A Functional Networks Softsensor For Flowing Bottomhole Pressures and Temperatures
in Multiphase Flow Production Wells. Presented at SPE Intelligent Energy Conference & Exhibition, Utrecht, The Netherlands. 1-3
April. SPE-167881-MS. https://doi.org/10.2118/167881-MS.
SPE-192354-MS 8

Papay, J. 1968.ATermelestechnologiai Parameterek Valtozasa a Gazlelepk Muvelese Soran, OGIL MUSZ, Tud, Kuzl., Budapest. pp. 267-
273.
Papp, I. 1979.Uj modszer foldgazok elteresi tenyezojenek szamitasara. In: Koolaj es Foldgaz, 345−47.
Standing, M.B. and Katz, D.L. 1942.Density of Natural Gases. Transactions of the AIME, 146(01):140-149.
https://doi.org/10.2118/942140-G
Wang, P. 2003. Development and Applications of Production Optimization Techniques for Petroleum Fields. Stanford University, Stanford
Xiang. 2005. The Corresponding-States Principle and its Practice. Elsevier Science.
Zadeh, L.A. 1965.Fuzzy Sets. Information and Control, 8: 338–353.