clustring analysis carbonate
clustring analysis carbonate
This paper was prepared for presentation at the Offshore Technology Conference held in Houston, Texas, USA, 6 – 9 May 2019.
This paper was selected for presentation by an OTC 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 Offshore Technology Conference and are subject to correction by the author(s). The material does not necessarily reflect any
position of the Offshore Technology Conference, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Offshore 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 OTC copyright.
Abstract
Facies prediction is an important step in reservoir characterization and modeling. Define a representative
reservoir model will enhance the process of reservoir development and will optimize the economic
strategies. The lack of data is a key issue in reservoir characterizations and therefore alternative approaches
have to be adopted to improve the process of reservoir characterization. In this research, clustering analysis
was implemented as a statistical solution to classify reservoir facies given well logs and core data in reservoir
from the south of Iraq.
In this research, data from a heterogeneous carbonate reservoir used. The data included well log records
such; GR, SP, Density, Neutron Porosity, Total Porosity, Resistivity, Induction, Shale Volume, Water
Saturation, along with porosity and permeability values from core analysis. These data were integrated and
analyzed through statistical tools to perform clustering analysis. The clustering analysis is an approach of
finding the similarities and differences between specific groups or points in order to classify them into
different classes. This concept was implemented by the use of R software, which is a quite powerful open
source tool for statistical studies with variety of functions and packages. Two different clustering algorithms,
K-mean approach and Calinski-Harabasz solution were used to classify reservoir facies based on the given
data.
The results of this research show that the reservoir facies can be predicted through different clustering
analysis when well logs records are given. K-means approach has predicted the optimal facies classification
to be five, while Calinski-Harabasz technique has identified three optimal reservoir facies. The difference
in facies classification between the two clustering analysis approaches is attributed to the two approaches
sensitivity or because of the high quality rocks in all the units of this well which makes it challenging to
identify the facies as all the layers have close reservoir properties.
Introduction
Efficient facies prediction and modeling is a vital step in reservoir characterization, which is essential in
reflecting reservoir heterogeneity and complexity. Reservoir flow modeling and performance is mainly
governed by defining the petrophysical properties and their spatial distribution through the reservoir.
Therefore, it is crucial to classify facies distribution as it first step to populate the petrophysical and the
2 OTC-29269-MS
other reservoir parameters effectively leading to building represented reservoir simulation models which are
imperative to implement optimum development strategies and economic plans (Teh et al. 2012). Clustering
analysis by R was used in this study to model the lithofacies in a carbonate reservoir given the well log
and core data.
Different statistical approaches and algorithms have been adopted to achieve lithofacies classification as it
is difficult to be done based on well logs because of complex heterogeneity of the carbonate facies structures.
A probabilistic neural network facies analysis was conducted to predict log facies by combining core
description and log analysis in a carbonate reservoir (Tang 2008). A gradient boosting classifier was also
adopted as a machine learning approach to provide facies classification. The classification was done based
on the wireline logging measurement by the use of small training set and amount of features (Bestagini et al.
2017). Seven multivariate statistical analysis algorithms of rock facies classification modelling, prediction,
and validation were studied and analyzed to define the optimum algorithms. The algorithms were Nave
Bayes Classifier (NBC), Logistic Boosting Regression (LogitBoost), Linear Discriminant Analysis (LDA),
Multinomial Logistic Regression (MLR), Tree-Based Classification Models (RPART), and Kernel Support
Vector Machine (KSVM), and the modeled rock facies based on known core and well log data in a sandstone
reservoir in West of Africa (Al-Mudhafar 2016). Random forest learning algorithm, is a tree-based classifier,
was used to anticipate the lithofacies given the wireline logs and stratigraphical interpretation. The seismic
facies classification and ranking the influence of each facies were achieved through this algorithm, which
is vital in reducing the amount of computations needed to define Lime and Shale facies in a Barnett shale
survey (Kim et al. 2018).
The key objective of this research is to integrate the given well log and core data in a statistical algorithm
in order to model and classify the facies and to define the spatial distribution. The used well log and core
data in this study are from a super-giant and very heterogonous carbonate reservoir that is located in South of
Iraq. In this research, clustering analysis was used to predict the facies through three approaches including
hierarchical agglomerative, partitioning, and model based. These approaches were applied on the given data
in R, which is a very powerful and influential open-source statistical computing language.
Mishrif is a quite complex and heterogeneous carbonate reservoir that is rich of light hydrocarbons. It has
a large distribution window of marine facies and sediment in a great thickness of more than 200 m. Rudist
reef and shoal facies limestones of the Mishrif reservoir compose an outstanding potential reservoir rocks.
In terms of stratigraphical sequence, it was divided three main oil units; MA, MB, and MC (Sang et al.
2017). However, other studies have divided Mishrif reservoir into five producing layers named MA, MB1-1,
MB1-2, MB2-1, and MB2-2). These layers found to be in communication in some parts of the reservoir
and they are described by a high total porosity. The main reservoir facies good oil bearing units which were
identified are: mud dominated Wackstone, grain dominated, bioclasticpelloidal and/or Pelloidal Grainstone
microfacies, Bondstone, rudist Packstone, coralline algal Wackstone–Packstone microfacies and dolomitic
Lime Mudstone – Wackstone microfacies (Rafed 2011). Given the above facies along with the available
well log and core data in a well in this reservoir, statistical approached to be adopted to predict the different
facies and Mishrif reservoir through the implementation of the clustering analysis by different models.
Lithofacies classification
Cluster analysis was adopted in this research as a statistical approach to model and predict the spatial
distribution of the reservoir lithofacises given well log data including; GR, SP, Density, Neutron Porosity,
Total Porosity, Resistivity, Induction, Shale Volume, Water Saturation, along with porosity and permeability
values from core analysis. The modeling process was implemented in R, which is an effective open source
4 OTC-29269-MS
tool for statistical computing and graphing which has a variety of packages and functions. The well log
shows that the majority of the units in this well have good quality and therefore, lithofacies classification
might be slightly challenging in this well as the difference in the petrophysical parameters will not be
significant. The available data for core description proposed the existence of five different facies as shown
in the figure below:
Well log with the characterized five facies from core description
As core data is expensive and only available in few wells, the key objective in this research is to define
a reliable statistical model to predict and model the lithofacies in this well and later to use this model in the
other wells to predict the facies when core data in unavailable.
Cluster Analysis
R software has broad variety of functions and packages for cluster analysis. Cluster analysis is a conventional
practice that is implemented to classify items and cases into different categories that also called as clusters.
It is also known as classification analysis or numerical taxonomy in which there is no previous information
regarding the cluster or the membership for any item (Clustering Analysis, 2018). The cluster process
is don based on the similarities in the data considering the information in the given data in order to
achieve groups or clusters have similar objective but totally different that the other clusters. There are
different types cluster including; well-separated, prototrype-based, graph-based, density-based, and shared
property (conceptual clusters), while there are few major clustering analysis techniques involving; K-
means (partitioning), Calinski-Harabasz, Agglomerative Hierarchical Clustering, and DBSCAN, which are
described below (Kumar & Steinbach, 2005).
OTC-29269-MS 5
Different methods of clustering similar sets of data (Kumar & Steinbach, 2005)
K-means (Partitioning)
K-means is one of the most common unsupervised learning algorithms that is usually used to analyzing the
well-known clustering problem. The K-means is considered as simple approach to classify a known data
into a specific number of groups, or clusters. Defining the K centroid for each cluster is basis of the K-
means approach. The placement of centroids must be precise as the different placement leads to different
outcomes (Cluster Analysis, 2018). Once centroids placement is done, the next step to specify the points
that fit in a specific data set and assign them to the nearest centroid. Then, a new K centroid to be calculated
in order to assigned as a possible center for the clusters and the points of data sets has to be associated to the
new centroid. The process is iterative and it will continue to be repeated until the location of the centroids
to be constant. In summary, the K-means algorithm target to reduce objective function, which is the squared
error function (Kumar & Steinbach, 2005). The objective function is shown below:
Where:
J: objective fuction
K: number of clusters
n: number of cases
x: case
c: centroid
The above objective function will be implemented by the use of R software to model and predict the
facies based on the given petrophysical data.
6 OTC-29269-MS
Calinski-Harabasz
The Calinski-Harabasz approach is a sample clustering technique which utilizes an index that represents
a ratio of a between cluster-means and a within-cluster sum of squares (Desgraupes, 2017). The Calinski-
Harabasz index can be described mathematically:
Where:
K: number of clusters
B(k) & W(k) : the between and the within cluster sums of square of partitions
Similar to K-means, this approach was implemented in this research by using R software and different
clusters were achieved which reflect different facies.
K-Means Partitioning
The K-Mean clustering has been used through R software and five vertical rock types (classes) sequences
along the whole depth interval of the well were identified. Figure 4, 5, and 6 show that the optimal classes
(Facies) are five that were identified based on the giving well log records. K-means approach predicted the
same number of facies as described in the core descriptions that enhance the reliability of this statistical
approach.
8 OTC-29269-MS
Calinski-Harabasz Classification
Only three classes, facies, were identified by the Calinski-Harabasz approach. This approach identified the
top and the bottom facies while combine the three facies in the middle, which were classified by the K-
Mean in to one class. This classification could be attributed to the lack of the sensitivity in this approach,
which was not enough to mimic reservoir facieses as this well has high quality rocks in all the units so the
differences are not significant. Figures 7, 8 and 9 depict the three identified facies by the Calinski-Harabasz
approach.
10 OTC-29269-MS
Finally, the predicted facies from both approaches were plotted against the given GR, permeability, and
porosity in or der to have an insight regarding the parameters, which has a direct impact on the clustering
process. Figure 9 and 10 depict that GR is a key parameters that control facies classification.
Conclusion
The different petrophysical properties form well logs along with core permeability and porosity have been
integrated through clustering analysis approach to predict and to model reservoir lithofacies. K-means has
been implemented as one of the clustering analysis technique and five different facies were identified. The
outcome of the K-means match facies classification from the given core description which enhance the
reliability of K-mean technique. The Calinski-Harabasz was adopted in this research and this approach
proposed three different reservoir facies. The different in the predicted number of facies can be attributed
to the high quality rocks in all reservoir units, which cause facies characterization to be challenging. In
summary, clustering analysis has been proven to be very efficient in facies classification, which is a key step
in reservoir models characterization that leads to improve reservoir models performance in development
plans and economic analysis.
References
AL-Mohammad, R. A. (2012). Depositional Environment and Petrophysical Properties Studyof Mishrif. Journal of
Basrah Researches.
Al-Mudhafar, W. J. (2016). Applied Geostatistical Reservoir Characterization in R: Review and Implementation of
Rock Facies Classification and Prediction Algorithms-Part I. Offshore Technology Conference. Houston: Offshore
Technology Conference.
Bestagini, P., Lipari, V., & Tubaro, S. (2017). A machine learning approach to facies classification using well logs. 2017
SEG International Exposition and Annual Meeting. Houston: Society of Exploration Geophysicists.
Cluster Analysis. (2018). Retrieved from statisticssolutions: https://www.statisticssolutions.com/directory-of-statistical-
analyses-cluster-analysis/
Clustering Analysis. (2018). Retrieved from Politecnico : https://home.deib.polimi.it/matteucc/Clustering/tutorial_html/
kmeans.html
Desgraupes, B. (2017). R software. Retrieved from Clustering Indices: https://cran.r-project.org/web/packages/clusterCrit/
vignettes/clusterCrit.pdf
OTC-29269-MS 13
Kim, Y., Hardisty, R., Torres, E., & Marfurt, K. J. (2018). Seismic facies classification using random forest algorithm.
Seismic facies classification using random forest algorithm. California: Society of Exploration Geophysicists.
Kumar, V., & Steinbach, M. (2005).
Sang, H., Lin, C., & Jiang, Y. (2017). Sequence stratigraphy and sedimentary study on Mishrif. International Symposium
on Resource Exploration and Environmental Science.
Tang, H. (2008). Improved Carbonate Reservoir Facies Classification Using Artificial Neural Network Method. Calgary,
Alberta: Petroleum Society of Canada.
Teh W, Willhite GP, Doveton JH (2012) Improved reservoir characterization in the Ogallah field using petrophysical
classifiers within electrofacies. Paper presented at the SPE improved oil recovery symposium, Tulsa, Oklahoma.
https://doi.org/10.2118/154341-MS