Gravity Map Production Using General Regression Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 02 | Feb -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 7
Gravity Map Production Using General Regression Neural Network
Nagi Ishag Mohammed 1, El Tahir Mohammed Hussein2, Adil Mohammed El Sinnari 3
1 National Ribat University, College of Graduate Studies and Scientific Research, Khartoum, Sudan
2Sudan University of Science and Technology, College of Engineering, Khartoum, Sudan
3Omdurman Islamic University, College of Engineering, Khartoum, Sudan
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Gravity map production has been improved by
using very complicated and expensive versions of geophysics
methods [1].
The main objective of this study is to evaluate the
ability of the Artificial Intelligent Techniques product gravity
map.
To achieve the objective of this study ArtificialNeural
Network architecture has been tested. This is the GRNN. The
GRNN model was trained with 301 patterns derived from
gravity map and satellite image. The maps were converting to
ASCII to generate the input part of the learning patterns. The
same rows were used to generate the output part. A
performance test session was carried out by applying the
trained models to the same training patterns and to 75 new
test patterns. The output results have been subjected to
statistical analysis.
Key Words: AI: Artificial Intelligent, ANN: Artificial Neural
Networks, ASCII: American Standard Code for Information
Interchange, GRNN: General Regression Neural Networks,
GRACE: Gravity Recovery and Climate Experiment.
.
1. INTRODUCTION
Gravity is a potential field, i.e., it is a force that acts
at a distance. The gravity method is a non-destructive
geophysical technique that measures differences in the
earth’s gravitational field at specific locations. It has found
numerous applications in engineering, environmental and
geothermal studies including locating voids, faults, buried
stream valleys, water table levels and geothermal heat
sources. The success of the gravity method depends on the
different earth materials having different bulk densities
(mass) that producevariationsinthemeasuredgravitational
field. These variations can then be interpretedbya varietyof
analytical and computers methods to determine the depth,
geometry and density that causesthegravityfieldvariations.
The conventional gravity map is sufficient very
costly and not satisfactory enough satisfy the criteria of the
research. While the gravity map presents a very rich
indicator in oil exploration, to overcome and solve this
problem. The research presents a satellite image for the
same area, and by using artificial intelligent techniques[1,2,
3].
The main objective of this study is to investigatethe
ability of artificial neural networks to product gravity map.
This will be achieved by training and testing an appropriate
ANN architecture with learning patterns generated from
satellite image and corresponding accurate values derived
from a satellite image.
To improve the opportunity to find oil, geologists
apply earth science to the search for oil. Many techniques
have been developed, based on indirect methods to view
the subsurface. Among the most important are:
(i) Seismology, which is the study of the sound
waves that bounce off buried rock layers. It
involves seismic surveys that are analyzed by
knowledgeable personnel [4].
(ii) Geological Mapping, which is used by
geologist to define possible reservoir shapes
or traps, due to the deformation in the rock
layer that contains hydrocarbons [5].
(iii) Educated guesses, which use physical geology
and seismic information as the base material to
guess where to drill [6].
To solve this problem:
a) High spectral space images will be enhanced
to highlight gravity issues.
b) An existing gravity map will be georeference
to bring the whole test data into a common
georeference.
c) Training and test patterns will be generated
from the satellite image and the existing
gravity map.

d) An ANN’s system will be built, specified and
trained for predicting gravity indicators
(map) based on previous experience.
e) The performance of the trained model will be
checked by applying it to a new set of test
data (patterns), whose actual outputs are
already known.
f) The results will be analyzed, discussed and
assessed.
2. STUDY AREA:
The experimental test carried out in this study was based on
test data collected and observed from satellite image of
Khartoum City (Figure 1).
Figure 1: Satellite Image for Khartoum City
The satellite image of Khartoum City (Figure 1)was
acquired by theAmericanIKONOSimaginingsystemin2005.
The spatial resolution of this image is one meter, while the
spectral resolution covers the visible (two bands) and the
infrared (one band) spectrum regions. Table 1 lists the
specifications of this image.
One of the most important set of the test data used
also in this study is a gravity map of Khartoum City (Figure
2). Earth's gravity measured by NASA's GRACE mission,
showing deviations from the theoretical gravity of an
idealized smooth Earth, the so-called earth ellipsoid. Red
shows the areas where gravity is stronger than the smooth,
standard value, and blue reveals areas where gravity is
weaker.
Some data has been derived from gravity map as
shown in figure 2.
Figure 2: Gravity map of Khartoum City
3. PROCEDURES:
A wide range of procedures were applied to meet
the objectives of this study. The test was carried out in three
phases. Phase one is concerned with image georeferencing,
image slicing and production of ASCII images. Inphasetwoa
set of training and test patterns were generate to build an
artificial neural network model to predict the gravitymapof
the test patterns based on the knowledge that was gained
during the training session. In phase three the performance
of the trained model was assessed.
4. RESULTS AND ANALYSIS:
The GRNN architecture was designed and trained
with 5000 patterns. GRNN achieved in less than 4 minutes.
The small error value is a quick indication that the shell was
able to converge with a minimum average error, and hence
reasonable activation state has been achieved by the
network. This indicates thattheGRNN looksmoresuitable in
resolving the research problem.
In order to assess the training results, the trained
model was applied to the same training patterns. The
predicted values were compared to the actual values
(derived from gravity map). Since the implicit relationships
between the input and output variables of the training
patterns are supposed to be well defined to GRNN model,
small residual values are expected to be obtained. Therefore
one can assess the model was well trained.
To assess the performance of the GRNN model for
predicting improved values based on input data obtained by
gravity map, the shell was applied to 116 new patterns (test
patterns) not included in the training session.

The predicted values were compared with original
gravity map. The absolute average of averages and the
standard deviation of averages for these residuals were also
computed. For the purpose of comparison and assessment,
the original gravity map was compared with predicted
values. The figures below explain pattern 6 (randomly) in
training session (gravity and predicted) and the difference
between two images. The black area means there are
difference between two images in the third figure. And the
white color means there is no difference between two
images.
Figure 3: The Gravity Pattern.
Figure 4: The Predicted Pattern.
Figure 5: The Difference between Two Patterns.
The histogram illustrate that there is no difference
between gravity image and predicted image for the same
area.
Figure 6: Histogram for Gravity Pattern.
Figure 7: Histogram for Predicted Pattern.
5. CONCLUSION:
In the present research work the research is
aiming to product gravity map using artificial intelligence
technique.
According to the test carried out in this study it can be
concluded that:

1. The GRNN model was able to process the whole
set of the training patterns (410 patterns), as well
as the test patterns (109 patterns).
2. The Artificial Neural Network is able to produce
Gravity Map.
ACKNOWLEDGEMENT
I am grateful to my supervisors Dr. El Tahir Mohammed
Hussein. And Dr. Adil El Sinnari who offered his academic
experience, both through his guidance and in discussion.
Without his continuous assistance and encouragement I
would not have been able to complete this work.
I am very thankful to all staff member of the National Ribat
University. My thanks are due to several of my colleagues
and friends for their spiritual support.
Finally special thanks to my family for their patience and
cooperation during the preparation period of this work.
REFERENCES
[1] http://www.minerals.er.usgs.gov, June 2016.
[2] http://www.nmariita@KENGEN.co.ke, March 2016.
[3] Jadhawar, Prashant Sopanrao, University of Adelaide,
Australian School of Petroleum, 2010.
[4] Landry, Todd J., Mapping and Two-Dimensional
Modeling of Basement and Smackover Topography
with Filtered Gravity Data in West-Central Mississippi
M.S., University of Louisiana at Lafayette, 2015.
[5] Wiese, David N., Optimizing Two Pairs of GRACE-like
Satellites for Recovering Temporal Gravity Variations ,
Ph.D., University of Colorado at Boulder, 2011.
[6] Walker, Christopher David, A gravityslideorigin forthe
Mormon Peak detachment: Re-examining the evidence
for extreme extension in the Mormon Mountains,
southeastern Nevada, U.S.A. Ph.D.,Columbia University,
2008.
BIOGRAPHIES
Nagi Ishag Mohammed
received the B.S.
degrees in Computer Science
from College of science and
Technology in 2002 and M.S. in
computer science from
university of Gezira in 2007
respectively. he stayed in
Jordanian Sudanese College
and Mashreq University as a
teacher of computer science
from 2004 until now.
El Tahir Mohammed Husein
received the Ph.D. in
Control EngineeringfromChina
University in, 1997. He stayed
in Sudan University of Science
and Technology as a teacher of
Control Engineering from1997
until now.
Adil Mohammed Ahmed
Elsinnari received the
B.S. M.S and Ph.D. in Survey
Engineering from University of
Khartoum 1985,1990and2004
respectively. he stayed in
University of Khartoum and
Omdurman Islamic University
as a teacher of Survey
Engineering from 1993 until
now.

Gravity Map Production Using General Regression Neural Network

More Related Content

What's hot (19)

Similar to Gravity Map Production Using General Regression Neural Network (9)

More from IRJET Journal (20)

Recently uploaded (20)

Gravity Map Production Using General Regression Neural Network