Poster Sample Slide

You might also like

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 1

Najam us saqib (H1714039) 3RD YEAR CATUNEANU MODEL (identification of sedimentary facies by well logs)

Department of Geology Materials and methods


University of Karachi Well logging data and cores
Core observation and description of this study are based
on core samples (longer than 141 m) obtained from three
ASAQUENCE ATRATIGRAPHY: wells(K611, K612, and K231) in the H-Zone of the Pearl
framework of genetically related stratigraphic facies geometries and their bounding surfaces used to determine depositional setting River Mouth Basin. This manual work was carried out by
experienced geologists and took substantial time and
effort to produce. All of these underground data, which
Abstract Taking K-successions of the H-Zone of the Pearl River Mouth Basin as a testing example, we used two kinds were ac-quired from the CNOOC Shenzhen Branch, are
of approaches to implement the microfacies identification. One is a direct identification, the other is an indirect available for researchers who meet the criteria of the
approach in which we conducted the lithofacies classification first and then identified the microfacies based on CNOOC Shenzhen Branch for access to confidential data.
previously estimated lithofacies. Both approaches were trained and checked by interpretations of experienced The data of the wireline logs are relatively rich and could
geologists from real subsurface core data. Multinomial logistic regression (MLR) and artificial neural network (ANN) cover most wells in the H-Zone. We chose a combination
were used in these two approaches as classification algorithms. Cross-validations were implement-ed. The source of three types of wireline logs, (1) acoustic delta time
data set was randomly divided into training subset and testing subset. Four models, namely, (DTE), (2)gamma ray (GR), and (3) density (RHOBE), which
MLR_direct,ANN_direct, MLR_indirect, and ANN_indirect, were trained with the training subset. The result of the is a com-mon choice to implement log-facies recognition
testing set shows that the direct approaches (MLR_direct and ANN_direct) perform relatively poor with a total (Saggaf andNebrija2000 ; Nebrija and Saggaf 2003). The
accuracy around 75%. While the indirect approaches (MLR_indirect and ANN_indirect) perform much better with a relationships between sedimentary microfacies and
total accuracy of around 89 and 82%,respectively. This indirect method is simple and reproducible, and it could lead wireline logs (DTE,GR, and RHOBE).
to a robust way of analyzing sedimentary microfacies of horizontal wells with little core data or even are almost Direct or indirect identification of microfacies
never cored while core data are available for nearby vertical wells. Multinomial logistic regression (MLR) and artificial neural
network (ANN) were used in these two approaches as
Introduction: Sedimentary microfacies play a significant role in the exploration and development of reservoirs,
classification algorithms (for more information about
especially in the prediction of remaining oil (Li et al.2015 ). With the growing de-mand for the accuracy of geological these two algorithms, see Appendix Part1). Cross-
researches in oilfields(denser well patterns, thinner vertical units, etc.), however, traditional manual identification of validations were implemented. The source data set was
microfacies has shown several deficiencies: overload working, low efficiency, and heavy dependence on geologists randomly divided into atraining subset (70%) and a testing
'ability and experience. Additionally, available cores and outcrops heavily required by traditional approaches are subset (30%). Four models, namely, MLR_direct,
usually limited. Thus, the search for more automated identification mechanisms linking sedimentary microfacies ANN_direct, MLR_indirect, and ANN_indirect, were
and readily available data sources (such as the logging data) is highly desirable (Ma et al.2000 ;LiandAnderson- trained with the training subset. Among these four
Sprecher2006 ). With the development of digital logging technology and computer science, the use of mathematical models, MLR_direct and ANN_direct are both direct
methods, such as statistical probability, artificial neural network, fuzzy logic, and so on, the automatic identification models. As for these direct models, we used the wireline
of sedimentary facies is becoming increasingly popular. The past decades have seen increasing numbers of studies logs along with the observed microfacies of the training
in sedimentary facies recognition with mathematical methods(Saggaf and Nebrija2000 ; Saggaf and Nebrija2003 data set to train the MLR and ANN models, which could be
;LiandAnderson-Sprecher2006 ; Tang and White2008 ). Yet, these studies mainly focused on the identification of used to distinguish the microfacies type directly. As for the
relatively larger scales of sedimentary facies, which could reveal sedimentary indirect models (i.e., theMLR_indirect and the
environments well, but are too large for microfacies description in reservoir characterization. There are also many ANN_indirect models), we used the wireline logs along
with the observed lithofacies of the training data set to
achievements in the field of lithologic facies or even geophysical parameters (Cuddy2000 ; Siripitayananon et
train the MLR and ANN models. These indirect models
al.2001 ).Identification of lithologic or geophysical parameters is less challenging because the relationship between
were used to identify the lithofacies first. Since
lithology (or physical properties) and log response is stronger, while these identifications did not go to sedimentary
sedimentary microfacies are combinations of specific
microfacies, which made them too microscopic for sedimentary mapping. In addition, there are also identification lithofacies, we combined the identified lithofacies into
methods (Ma et al.2000 ;Lied al.2011 ) based on logging curve morphology. These identifications indeed focus on microfacies after the lithofacies were identified by the
sedimentary microfacies, but they rely too much on a certain well logging curve (usually GR),which makes the indirect models. The combinations from lithofacies to
results less reliable. Taking the wave-dominated K-successions of the Pearl River Mouth Basin as a testing example, microfacies must follow some rules. Under fundamental
based on the existing researches above, we aim to establish a more stable method to model the microfacies as a geologic principles and the geologists 'experience, we
function of well logs to predict them in non-measured intervals and/or other wells, which can not only meet the continued to modify the combination rules until the
accuracy need of fine reservoir description but also reflect depositional features in an appropriate scale. identified results match the observed results well. The
detailed combination rules are shown in AppendixPart2. R
programming language and RStudio software were used in
this study.

Results: The training data trained models perform differently on the testing data set. The results show that the direct approaches perform relatively
poor. Both the ANN_direct model and the MLR_direct model were gained with a total accuracy below 80% (77.8 and75.9%, respectively). As for the
ANN_direct model, the microfacies of OB was identified with an accuracy of about87.9% and the microfacies of OM is about 71.4%, while the
microfacies of OS is only 57.1%. As for the MLR_direct model, the microfacies of OB was identified with an accuracy of about91% and the microfacies of
OM is about 86%, while the microfacies of OS is only 35.7% (Fig.5).In contrast, the results of the indirect classification approaches perform much better
(Fig.6). Especially, the lithofacies estimated by the MLR_indirect model and the ANN_indirect model resemble the observed lithofacies of the testing
data set very well(Table2). Many classes reach the accuracy of 100%, while some individual class (i.e., Fp) is of low accuracy. The overall accuracy of the
MLR_indirect model and the ANN_indirect model is 94.4and 96.2%, respectively. Under certain rules, these estimatedlithofacies are combined into
microfacies. For the final estimation result of microfacies, the ANN_indirect model and the MLR_indirect model reach the overall accuracy of about
88.9and 81.5%, respectively. As for the ANN_indirect model, the microfacies of OB was identified with an accuracy of about87.9% and the microfacies
of OM is about 71.4%, while the microfacies of OS even reaches 100%. As for the MLR_indirect model, the microfacies of OB was identified with an
accuracy of about 87.9% and the microfacies of OM and O Bare of both more than 71%.

You might also like