Professional Documents
Culture Documents
IPTC-20208-MS Comparative Analysis Between Different Artificial Based Models For Predicting Static Poisson's Ratio of Sandstone Formations
IPTC-20208-MS Comparative Analysis Between Different Artificial Based Models For Predicting Static Poisson's Ratio of Sandstone Formations
IPTC-20208-MS Comparative Analysis Between Different Artificial Based Models For Predicting Static Poisson's Ratio of Sandstone Formations
This paper was prepared for presentation at the International Petroleum Technology Conference held in Dhahran, Saudi Arabia, 13 – 15 January 2020.
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
The mechanical behavior of the rocks can greatly assist in optimizing the drilling operation and well
completion design. This behavior can be expressed in terms of Young’s modulus and Poisson’s ratio.
Reliable Poisson’s ratio values can be estimated experimentally from core measurements however this
method consumes time and economically ineffective.
This study involved the development of two models using neural networks (ANN) and fuzzy logic to
estimate static Poisson’s ratio (PRstatic) of sandstone rocks based on the conventional well-log data including
bulk density and sonic log data. The models are developed using 692 of actual data core data and the
corresponding logging data. The models are optimized after several runs of the different combinations of
the available tuning parameters.
The results showed that the neural network model outperformed the model developed using the fuzzy
logic tool and yielded a great match with correlation coefficient (R) of 0.98 and AAPE of 1.5% between
the predicted and measured PRstatic values. The developed ANN-based model is then validated using unseen
data from another well within the field to estimate PRstatic over a certain interval. The validation process
results showed a significant agreement with correlation coefficient (R) of 0.95 between the predicted PRstatic
values and the actual measured ones. The results demonstrated the ability of the developed model to provide
a continuous profile of static Poisson’s ratio (PRstatic) whenever the petrophysical logging data are available.
Introduction
Poisson's ratio is a geomechanical property for the material and it is defined as the ratio of the lateral
increment to the longitudinal decrement under compression (Tutuncu and Sharma, 1992). Hence, negative
of the ratio of lateral or transverse strain to axial or longitudinal strain represents the static Poisson's ratio
measurement and that strain should be resulted from uniaxial force only (Kumar, 1976). Geo-mechanical
earth models are one of the tools used to represent the in-situ state of rock. The development of such models
depends on the in-situ stresses encountered within a formation, which can be estimated using the values
of its elastic parameters, Poisson’s ratio, and Young’s modulus. These parameters are very important in
describing the elastic behavior of rock. The materials that show a linear elastic behavior, found to have
2 IPTC-20208-MS
Poisson's ratio values range from -1 to 0.5 (Truesdell, 1984). However, for typical rocks, found that the
value of Poisson's ratio ranges from 0.1 to 0.4 as the lower values for the soft formations and the higher
values for the hard rocks (Gercek, 2007).
Several factors have an impact on the Poisson's ratio as rock lithology, density, porosity and pore
structure, consolidation, fluid saturation, and formation temperature (Kumar, 1976; Phani, 2008; Ameen
et al., 2009; Wang et al., 2009; Al-Anazi and Gates, 2010). The rock petrophysical and geomechanical
parameters are so critical for designing the well drilling and completion programs, and geomechanical earth
(1)
(2)
Zhang and Bentley (2005) showed that the difference between the two values for the static and dynamic
Poisson's ratio resulted from the variation of strain values for the two methods of measurements. The lab
measurements required core extraction from the well and measuring the parameters in conditions as same
as the reservoir and therefore it is so expensive and time consuming (Abdulraheem A., 2019). Alternatively,
correlations are established to get the static Poisson's ratio from the well logs (Gatens et al., 1990; Kumar
et al., 2003; Shalabi et al., 2007; Phani, 2008; Tariq et al., 2016; Elkatatny et al., 2018). However, such
correlations have many limitations that should be considered when using them. (Abdulraheem et al., 2009;
Ameen et al., 2009; Badrouchi et al., 2019). Artificial intelligence has been presented as an alternative way
for developing models which have the ability of dealing with problems with high degree of complexity
(Nagy, K. and Hajrizi, 2018; Agwu et al., 2018). From recent literature, several studies were performed
to predict Poisson's ratio using AI techniques (Tariq et al., 2019; Abdulraheem, 2019; Badrouchi et al.,
2019; Elkatatny et al., 2019). However, most of them are related to carbonate rocks. Gowida et al. (2019)
predicted the static Poisson’s ratio for sandstone rocks using well-log data such as bulk density and sonic
log using ANN and showed that the PRstatic can be estimated with high accuracy using well-log data as
inputs. Therefore the objective of this work is to present a comparative analysis between using ANN and
ANFIS for building predictive models for PRstatic and compare their performance based on the accuracy of
the results obtained.
fundamental units called neurons for processing the data (Niculescu, 2003; Razi and Naderi, 2009). These
neurons are connected within the systems using connections to imitate the biological structure of the human
neural system. Each connection is assigned with weights and biases defined during the training process
(Rao and Ramamurti, 1993). ANN is basically trained using different algorithms to analyze the system and
optimize the weights and biases (Nakamoto, 2017). Among these algorithms, back-propagation algorithm
was suggested as an outperforming learning technique (Hinton et al., 2006; Yagiz et al., 2012). A typical
structure of ANN model involves three main types of layers namely; input, hidden, and output layers. Input
Data Description
In this study, both laboratory measurements (core data) and wireline-logged data from drilled wells were
used for training the models developed by different AI tools (ANN, ANFIS).
Table 1—Statistical parameters of the obtained core data and well-log data.
Table 2—The tested options of the tuning parameters for ANN and ANFIS.
Number of
1-4 1
Hidden layers
Number of Neurons
1-30 20
in Each Layer
tansig logsig
trainlm trainr
Tansig (Tans-
Transfer Function trainbr trainc
Sigmoidal)
traingd traingda
The developed models are then compared to select the model with the best performance in terms of R
and AAPE. Based on the obtained results, the ANN-based model outperformed the ANFIS-based model
as it yielded the highest R of 0.97 and the lowest AAPE of 1.6 % as shown in Figs. 2a and 2b. Figs. 3
and 4 show cross-plots between the predicted and measured PRstatic values for training and testing processes
using the ANN-based and ANFIS-based models. The results of the training and testing for both methods
are summarized in Table. 3.
Figure 2a—Comaprison between the predicted and actual PRstatic for the testing data using the ANFIS-based model.
Figure 2b—Comaprison between the predicted and actual PRstatic for the testing data using the ANFIS-based model.
IPTC-20208-MS 7
R 0.97 0.98
ANN
AAPE (%) 1.67 1.54
R 0.92 0.94
ANFIS
AAPE (%) 2.72 3.35
Model Validation
The developed ANN-based model is then validated using validation data form another well within the area
(These data are not included in the model development process). The validation data involved conventional
well log data (RHOB, ∆tcomp and ∆tshear) and five data points of core measurements within an interval of 550
ft. of sandstone formation. The obtained well-log data are used to feed the network to get a continuous profile
of the PRstatic then the results would be compared with the actual core measurements to check the competency
of the prediction performance of the developed ANN-based model. Fig. 5 shows that the predicted results
of the ANN-based model significantly match with the core measurements (R = 0.95, AAPE = 3.5%)
8 IPTC-20208-MS
Conclusions
Comparative analysis has been performed to evaluate the prediction performance of static Poisson’s ratio
using two artificial intelligence tools namely; neural networks and fuzzy logic, leading to the following
findings:
1. The developed ANN-based model showed outperformance over the ANFIS-based model in predicting
PRstatic using the well-log data as inputs.
2. The developed ANN-based model can provide a continuous profile of PRstatic whenever the well-log-
data are available, with R of 0.98 and AAPE of 1.54 % between the predicted and measured PRstatic
values.
3. The validation process of the developed ANN-based model using unseen data from another well in
the same area demonstrated its great prediction performance of PRstatic with R of 0.95 between the
predicted and measured PRstatic values.
4. The developed ANN-based model is considered a timely effective tool to estimate PRstatic values of
sandstone rocks, especially when core data are not available.
Nomenclature
AI Artificial Intelligence
AAPE Average absolute percentage error
ANN Artificial neural network
Tansig Hyperbolic tangent sigmoid transfer function
Hardlim Hard-limit transfer function
Logsig Log-sigmoid transfer function
Pure-linear Linear transfer function
Elliotsig Elliot symmetric sigmoid transfer function
Tribas Triangular basis transfer function
Satlin Saturating linear transfer function
Radbas Radial basis transfer function
Trainlm Levenberg-Marquardt backpropagation
Trainr Random order incremental training with learning functions
Traingda Gradient descent with adaptive learning rule backpropagation
Trainc Cyclical order incremental update
Trainbr Bayesian Regularization backpropagation
Traingd Gradient descent backpropagation
IPTC-20208-MS 9
References
1. Abdulraheem, A. 2019, March 22. Prediction of Poisson’s Ratio for Carbonate Rocks Using
ANN and Fuzzy Logic Type-2 Approaches. Paper presented at the International Petroleum
Technology Conference, Beijing, China, 26-28 March. IPTC-19365-MS. https://doi.org/10.2523/
IPTC-19365-MS
2. Abdulraheem, A., M. Ahmed, A. Vantala, and T. Parvez. 2009. Prediction of rock mechanical
parameters for hydrocarbon reservoirs using different artificial intelligence techniques. Paper
15. Chandrasegar, T.; Vignesh, M.; Balaji, R. Data Analysis Using Box and Whisker Plot for Lung
Cancer. In Proceedings of the Innovations in Power and Advanced Computing Technologies (i-
PACT) Conference, Vellore, India, 21–22 April 2017.
16. Chang, C., Zoback, M.D. and Khaksar, A., 2006. Empirical relations between rock strength and
physical properties in sedimentary rocks. Journal of Petroleum Science and Engineering, 51(3-4):
223-237. https://doi.org/10.1016/j.petrol.2006.01.003.
17. Dawson, R. How Significant Is A Boxplot Outlier? J. Stat. Educ. 2011, 19. [CrossRef]
33. Lippman RP. 1987. An introduction to computing with neural nets. IEEE ASSP Magazine 4:
pages 4-22. https://doi.org/10.1109/massp.1987.1165576.
34. Nakamoto P. Neural networks and deep learning: deep learning explained to your granny a visual
introduction for beginners who want to make their own deep learning neural network (machine
learning), 2017, CreateSpace Independent Publishing Platform.
35. Nes, O., Fjær, E., Tronvoll, J., Kristiansen, T. G., and Horsrud, P. 2012. Drilling Time Reduction
Through an Integrated Rock Mechanics Analysis." ASME. J. Energy Resour. Technol. 134(3)
49. Truesdell, C., 1984. Timoshenko's History of Strength of Materials (1953), in: An Idiot's
Fugitive Essays on Science. Springer New York, New York, NY, pp. 251–253. https://
doi.org/10.1007/978-1-4613-8185-3_30
50. Tutuncu, A.N., Sharma, M.M. 1992. Relating Static and Ultrasonic Laboratory Measurements
to Acoustic Log Measurements in Tight Gas Sands. Paper presented in the SPE Annual
Technical Conference and Exhibition, Washington, D.C., 4-7 October. SPE-24689-MS. https://
doi.org/10.2118/24689-MS.