You are on page 1of 188

DOKUZ EYLÜL UNIVERSITY

GRADUATE SCHOOL OF NATURAL AND APPLIED SCIENCES

USE OF ARTIFICIAL INTELLIGENCE


OPTIMIZATION ALGORITHMS IN THE
INVERSION OF GEOPHYSICAL DATA

by
Seçil TURAN KARAOĞLAN

February, 2023
İZMİR
USE OF ARTIFICIAL INTELLIGENCE
OPTIMIZATION ALGORITHMS IN THE
INVERSION OF GEOPHYSICAL DATA

A Thesis Submitted to the


Graduate School of Natural and Applied Sciences of Dokuz Eylül University
In Partial Fulfillment of the Requirements for the Degree of Doctor of
Philosophy in Geophysical Engineering, Geophysical Engineering Program

by
Seçil TURAN KARAOĞLAN

February, 2023
İZMİR
Ph.D. THESIS EXAMINATION RESULT FORM

We have read the thesis entitled “USE OF ARTIFICIAL INTELLIGENCE


OPTIMIZATION ALGORITHMS IN THE INVERSION OF GEOPHYSICAL
DATA” completed by SEÇİL TURAN KARAOĞLAN under supervision of PROF.
DR. GÖKHAN GÖKTÜRKLER and we certify that in our opinion it is fully
adequate, in scope and in quality, as a thesis for the degree of Doctor of Philosophy.

Prof. Dr. Gökhan GÖKTÜRKLER

Supervisor

Prof. Dr. Aybars UĞUR Assist. Prof. Dr. Petek SINDIRGI

Thesis Committee Member Thesis Committee Member

Prof. Dr. Ali AYDIN Prof. Dr. Çağlayan BALKAYA

Examining Committee Member Examining Committee Member

Prof. Dr. OKAN FISTIKOĞLU


Director
Graduate School of Natural and Applied Sciences

ii
ACKNOWLEDGEMENTS

First and foremost, I would like to thank my PhD advisor, Prof. Dr. Gökhan
Göktürkler, who has been a great mentor throughout my work with his invaluable
guidance, advice, and essential support. I am grateful to my thesis advisor for his
immense patience and trust, especially during difficult times. His guidance helped me
throughout the research and writing of this thesis. I am honored to have had such a
great supervisor.

My special thanks to the members of my thesis committee, and I would also like to
thank Prof. Dr. Aybars Uğur for enlightening me with his valuable knowledge about
my research studies, and Assist. Prof. Dr. Petek Sındırgı for her comments and
feedback.

I would like to thank Dr. Xin-She Yang for the codes used in the thesis study. Also,
my special thanks to Prof. Dr. Çağlayan Balkaya for his support and useful advice on
programming with MATLAB, and MSc Sanam Hosseinzadeh, my colleague, for her
help and support.

Finally, I would like to thank my spouse, Dr. Mustafa Umut KARAOĞLAN, who
supported me during my thesis study with a lot of patience, motivation, and empathy.
I am also very grateful to my mother whom I am proud to be her daughter, my father
whom I believe is still watching me, and my dear sister for her strong support and
encouragement, and my nephews who are the joy of my life.

Seçil TURAN KARAOĞLAN

iii
USE OF ARTIFICIAL INTELLIGENCE OPTIMIZATION ALGORITHMS
IN THE INVERSION OF GEOPHYSICAL DATA

ABSTRACT

Traditional derivative-based and metaheuristic methods are currently used in


inversion studies of geophysical data with various geophysical research methods such
as self-potential and gravity. Due to the drawbacks of traditional derivative-based
methods, metaheuristic methods, which have higher computational time but reach the
global solution more effectively, play an important role in the inversion of geophysical
data.

In this study, we investigated the application of the Cuckoo Search Algorithm and
the Bat Algorithm for the inversion of self-potential and gravity data, which are widely
used in geophysics. The inversions of self-potential and gravity field data were
performed respectively for each application of the metaheuristic methods. Moreover,
the histograms of the relative frequency changes by the Metropolis-Hasting Algorithm
for the inversion results from synthetic and field data sets are presented. Finally, in
addition to the metaheuristic methods used in the inversion studies, a new
metaheuristic method called the "Perturbation-Based Evolutionary Algorithm" was
developed for the inversion of geophysical data, and the performance of the new
algorithm was evaluated using various test functions.

The obtained inversion results show that the Cuckoo Search Algorithm and the Bat
Algorithm can be effectively used for the inversion of self-potential and gravity field
data, since the values of the model parameters and the rms results are consistent with
the literature, and the frequency and distribution of the model parameters are consistent
with the iteration number. Additionally, the proposed Perturbation Based Evolutionary
Algorithm can be successfully applied to the inversion of geophysical data.

Keywords: Self-potential, gravity, cuckoo search algorithm, bat algorithm,


perturbation based evolutionary algorithm, inversion, geophysics

iv
YAPAY ZEKA OPTİMİZASYON ALGORİTMALARININ JEOFİZİK
VERİLERİN TERS ÇÖZÜMÜNDE KULLANILMASI

ÖZ

Doğal-Potansiyel ve gravite gibi çeşitli jeofizik yöntemleriyle jeofizik verilerinin


ters çözümü için günümüzde geleneksel türev tabanlı ve metasezgisel yöntemler
kullanılmaktadır. Türev tabanlı geleneksel yöntemlerin dezavantajları sebebiyle daha
yüksek hesaplama zamanına sahip olan fakat global çözüme daha efektif ulaşan
metasezgisel yöntemler jeofizik verilerinin ters çözümünde önemli yer tutmaktadır.

Bu çalışmada, jeofizikte yaygın kullanım alanına sahip doğal-potansiyel ve gravite


verilerinin ters çözümü için Guguk Kuşu Arama Algoritması ve Yarasa
Algoritmasının kullanımı incelenmiştir. Her bir metasezgisel yöntemin uygulamasında
sırasıyla doğal-potansiyel ve gravite verilerinin ters çözümü gerçekleştirilmiştir.
Ayrıca sentetik ve arazi verileri ile elde edilen ters çözüm sonuçları için Metropolis-
Hasting Algoritması ile rölatif frekans değişimlerinin histogram grafikleri
sunulmuştur. Son olarak, ters çözüm çalışmalarında uygulanan metasezgisel
yöntemlerin dışında jeofizik verilerinin ters çözümü için yeni bir metasezgisel
algoritma olan “Pertürbasyon Temelli Evrimsel Algoritma” geliştirilmiş ve çeşitli test
fonksiyonları kullanılarak bu algoritmanın performansı değerlendirilmiştir.

Elde edilen sonuçlara göre doğal-potansiyel ve gravite verilerinin ters çözümünde


Guguk Kuşu Arama Algoritması ve Yarasa Algoritmasının, literatürdeki sonuçlarla
tutarlı parametre ve rms değerleri vermesi ve iterasyon sayısına bağlı model parametre
değerlerinin değişimiyle frekansının uygun olması sebebiyle efektif olarak
kullanılabileceği görülmüştür. Ayrıca önerilen Pertürbasyon Temelli Evrimsel
Algoritmanın da jeofizik verilerin ters çözümüne uyarlanması halinde başarılı bir
şekilde kullanılabileceği anlaşılmıştır.

Anahtar Kelimeler: Doğal-potansiyel, gravite, guguk kuşu arama algoritması, yarasa


algoritması, pertürbasyon temelli evrimsel algoritma, ters çözüm, jeofizik

v
CONTENTS

Page
THESIS EXAMINATION RESULT FORM .............................................................. ii
ACKNOWLEDGEMENTS . ...................................................................................... iii
ABSTRACT ................................................................................................................ iv
ÖZ ................................................................................................................................ v
LIST OF FIGURES .................................................................................................... ix
LIST OF TABLES .................................................................................................... xiv

CHAPTER 1 - INTRODUCTION ............................................................................ 1

1.1 Motivation of Thesis ......................................................................................... 1


1.2 Thesis Objectives .............................................................................................. 3
1.3 Structure of Thesis............................................................................................. 5

CHAPTER 2 - AN OVERVIEW OF METAHEURISTIC METHODS IN


GEOPHYSICS ........................................................................................................... 7

2.1 Introduction to Metaheuristics........................................................................... 7


2.2 Classification of Metaheuristics ........................................................................ 9
2.3 Genetic Algorithm ........................................................................................... 11
2.4 Paricle Swarm Optimization ........................................................................... 14
2.5 Differential Evolution Algorithm .................................................................... 17
2.6 Simulated Annealing ....................................................................................... 21

CHAPTER 3 - LITERATURE REVIEW OF GEOPHYSICAL METHODS


BASED ON METAHEURISTICS .......................................................................... 24

3.1 Genetic Algorithm Applications ..................................................................... 24


3.2 Partical Swarm Optimization Applications ..................................................... 31
3.3 Simulated Annealing Applications .................................................................. 40
3.4 Differential Evolution Applications ................................................................ 45

vi
3.5 Applications of Other Metaheuristic Algorithms ............................................ 49

CHAPTER 4 - CUCKOO SEARCH ALGORITHM FOR SELF-POTENTIAL


AND GRAVITY ANOMALIES ............................................................................. 53

4.1 Inversion of SP Anomalies with Cuckoo Search Algorithm............................57


4.1.1 Mathematical Model of Self-Potential Anomaly......................................57
4.1.2 Parameter Tuning of Self-Potential Data for Cuckoo Search Algorithm...58
4.1.3 Numerical Tests and Metropolis Hastings Algorithm for Self-Potential...62
4.1.4 Tests with Theoretical Data for Self-Potential..........................................62
4.1.5 Tests with Field Data for Self-Potential...................................................69
4.1.5.1 Bavarian Woods Anomaly (Germany) ............................................69
4.1.5.2 Süleymanköy Anomaly (Türkiye)....................................................70
4.1.5.3 Malachite Mine Anomaly (USA).....................................................71
4.1.5.4 KTB Borehole Anomaly (Germany)................................................72
4.1.5.5 Uncertainty Analysis of Field Data Sets...........................................77
4.2 Inversion of Gravity Anomalies with Cuckoo Search Algorithm....................79
4.2.1 Mathematical Model of Gravity Anomaly...............................................79
4.2.2 Parameter Tuning of Gravity Data for Cuckoo Search Anomaly............80
4.2.3 Tests with Theoretical Data for Gravity...................................................82
4.2.4 Tests with Field Data for Gravity.............................................................87
4.2.4.1 Camaguey Anomaly (Cuba) ............................................................87
4.2.4.2 Quebec Anomaly (Canada) .............................................................88
4.3 Results and Discussions of Inversion Studies using CSA...............................91

CHAPTER 5 - BAT ALGORITHM FOR SELF-POTENTIAL AND GRAVITY


ANOMALIES ........................................................................................................... 93

5.1 Echolocation Characteristics of Bats................................................................93


5.2 Local Search Characteristics of Bat Algorithm...............................................94
5.3 Inversion of SP Anomalies with Bat Algorithm..............................................96
5.3.1 Parameter Tuning of Self-Potential Data for Bat Algorithm....................96

vii
5.3.2 Numerical Tests with Theoretical Data for Self-Potential.......................99
5.3.3 Tests with Field Data for Self-Potential.................................................106
5.3.3.1 Bavarian Woods Anomaly (Germany) ..........................................106
5.3.3.2 Malachite Mine Anomaly (USA) ..................................................107
5.3.3.3 Süleymanköy Anomaly (Türkiye)..................................................109
5.3.3.4 KTB Borehole Anomaly (Germany) .............................................111
5.4 Inversion of Gravity Anomalies with Bat Algorithm....................................114
5.4.1 Parameter Tuning of Gravity Data for Bat Algorithm...........................115
5.4.2 Numerical Tests with Theoretical Data for Gravity...............................118
5.4.3 Tests with Field Data for Gravity...........................................................123
5.4.3.1 Camaguey Anomaly (Cuba) ..........................................................123
5.4.3.2 Quebec Anomaly (Canada) ............................................................125
5. 5 Results and Discussions of Inversion Studies using BA...............................127

CHAPTER 6 - A NEW APPROACH: PERTURBATION-BASED


EVOLUTIONARY ALGORITHM ...................................................................... 129

6.1 Characteristics of the Perturbation-Based Evolution Algorithm.....................129


6.2 Perturbation-Based Evolutionary Algorithm with Sphere Test Function......133
6.3 Perturbation-Based Evolutionary Algorithm with Rosenbrock’s Test
Function...........................................................................................................134
6.4 Perturbation-Based Evolutionary Algorithm with Rastrigin’s Test
Function...........................................................................................................135
6.5 Perturbation-Based Evolutionary Algorithm with Himmelblau's Test
Function...........................................................................................................136
6.6 Perturbation-Based Evolutionary Algorithm with Beale Test Function........137

CHAPTER 7 - CONCLUSION ............................................................................. 139

REFERENCES ....................................................................................................... 144

viii
LIST OF FIGURES

Page
Figure 2.1 Pseudo code of GA (Modified from Yang, 2014) .................................... 13
Figure 2.2 Pseudo code of PSO (Modified from Yang, 2014)................................... 17
Figure 2.3 Pseudo code of DEA (Modified from Yang, 2014) .................................. 20
Figure 2.4 Pseudo code of SA (Modified from Yang, 2014) ..................................... 23
Figure 4.1 Flowchart of the cuckoo search algorithm (Turan-Karaoğlan and
Göktürkler, 2021) ...................................................................................... 55
Figure 4.2 Illustration of the primary geometrical bodies. (a) semi-infinite vertical
cylinder, (b) infinitely long horizontal cylinder and (c) sphere (Turan-
Karaoğlan and Göktürkler, 2021) .............................................................. 58
Figure 4.3 Synthetic SP anomaly for an infinitely long horizontal cylinder (Turan-
Karaoğlan and Göktürkler, 2021) .............................................................. 59
Figure 4.4 Change of the rms values with respect to Np for: (a) ρa =0.05, (b) ρa =0.15,
(c) ρa =0.25, (d) ρa =0.35 and (e) ρa =0.45. .............................................. 61
Figure 4.5 Change of the error energy during the solutions based on the theoretical
data (a) without noise and (c) with noise for 300 iterations, and comparison
of observed and calculated anomaly results for (b) noise-free and (d) noisy
SP data. (Turan-Karaoğlan and Göktürkler, 2021).................................... 63
Figure 4.6 Change of the model parameters (a-e) noise-free and (f-j) noisy theoretical
data sets according to iteration number. (Turan-Karaoğlan and Göktürkler,
2021) .......................................................................................................... 64
Figure 4.7 Error energy maps of SP synthetic noise-free data for model parameters
using CSA .................................................................................................. 65
Figure 4.8 Histograms via the M-H algorithm for (a-e) the noise-free and (f-j) noisy
data sets. The parameters obtained by the CSA are also indicated on each
histogram. (Turan-Karaoğlan and Göktürkler, 2021) ................................ 66
Figure 4.9 Error energy (a) and inversion result of multiple synthetic data for 600
iterations .................................................................................................... 67
Figure 4.10 Change of the model parameters (a-e) Source-1 and (f-j) Source-2 for
multiple synthetic data sets according to iteration number. ...................... 68

ix
Figure 4.11 The results from the Bavarian woods, Germany SP data set. (a) change of
the error energy and (b) the comparison of the observed and calculated
anomalies. (Turan-Karaoglan and Gokturkler, 2021) ............................... 70
Figure 4.12 The results from the Süleymanköy (Türkiye) SP data set. (a) change of the
error energy and (b) the comparison of the field anomaly and model
response. (Turan-Karaoğlan and Göktürkler, 2021).................................. 71
Figure 4.13 The results from the Malachite mine (Colorado, USA) SP data set. (a)
change of the error energy and (b) the comparison of the observed and
calculated anomalies. (Turan-Karaoğlan and Göktürkler, 2021) .............. 72
Figure 4.14 The results from the KTB Borehole (Germany) SP data set using vertical
cylinder model (Case-1). (a) change of the error energy and (b) the
comparison of the observed and calculated anomalies .............................. 73
Figure 4.15 Illustration of the inclined sheet geometry (Modified from Murty
Satyanarayana and Haricharan, 1985) ....................................................... 74
Figure 4.16 The results of CSA from the KTB Borehole (Germany) SP data set using
inclined sheet model (Case-2). (a) change of the error energy and (b) the
comparison of the observed and calculated anomalies .............................. 75
Figure 4.17 Subsurface structure for multiple inclined sheet model of the KTB
Borehole (Germany) according to the calculated anomalies by CSA ....... 77
Figure 4.18 Histograms via the M-H algorithm for the field data sets. (a-e) The
Bavarian, (f-j) Süleymanköy and (k-o) Malachite mine anomalies. The
estimated parameters by the CSA are also indicated on each histogram.
(Turan-Karaoğlan and Göktürkler, 2021) .................................................. 78
Figure 4.19 The anomalies of various bodies as sphere, infinitely long horizontal
cylinder, and semi-infinite vertical cylinder with their physical models
(Turan-Karaoğlan and Göktürkler, 2022) .................................................. 80
Figure 4.20 Inversion result of the noise-free data for 300 iteration (a) and error energy
according to the generation number (b) with CSA. (Turan-Karaoğlan and
Göktürkler, 2022) ...................................................................................... 82
Figure 4.21 Convergence of the model parameters for noise-free theoretical data
according to the generation number (a-e) (Turan-Karaoğlan and Göktürkler,
2022) .......................................................................................................... 83

x
Figure 4.22 Inversion result of the noisy data for 300 iteration (a) and error energy
according to the generation number (b) with CSA. (Turan-Karaoğlan and
Göktürkler, 2022) ...................................................................................... 84
Figure 4.23 Convergence of the model parameters for noisy theoretical data according
to the generation number (a-e). (Turan-Karaoğlan and Göktürkler, 2022) 84
Figure 4.24 Error energy maps of gravity synthetic noise-free data for model
parameters using CSA ............................................................................... 85
Figure 4.25 Histograms of relative frequency of noise-free and noisy SP theoretical
data by M-H algorithm. (Turan-Karaoğlan and Göktürkler, 2022) ........... 86
Figure 4.26 Inversion result of the Camaguey field anomaly for 300 iteration (a) and
error energy according to the generation number (b) with CSA. (Turan-
Karaoğlan and Göktürkler, 2022) .............................................................. 88
Figure 4.27 Convergence of the model parameters for Camaguey field anomaly
according to the generation number (a-e) (Turan-Karaoğlan and Göktürkler,
2022) .......................................................................................................... 88
Figure 4.28 Inversion result of the Quebec field anomaly for 300 iteration (a) and error
energy according to the generation number (b) with CSA. (Turan-Karaoğlan
and Göktürkler, 2022)................................................................................ 89
Figure 4.29 Convergence of the model parameters for Quebec field anomaly according
to the generation number (a-e). (Turan-Karaoğlan and Göktürkler, 2022) 90
Figure 4.30 Histograms of relative frequency of Camaguey and Quebec field data by
M-H algorithm. (Turan-Karaoğlan and Göktürkler, 2022) ....................... 90
Figure 5.1 Flowchart of the Bat Algorithm................................................................ 95
Figure 5.2 Inversion result of the SP noise-free data for 300 iteration (a) and error
energy according to the iteration (b) with Bat Algorithm ....................... 100
Figure 5.3 Convergence of the model parameters for SP noise-free theoretical data
according to the iteration number (a-e) ................................................... 100
Figure 5.4 Inversion result of the SP noisy data for 300 iteration (a) and error energy
according to the iteration (b) with Bat Algorithm ................................... 101
Figure 5.5 Convergence of the model parameters for SP noisy theoretical data
according to the iteration number (a-e) ................................................... 101

xi
Figure 5.6 Error energy maps of SP synthetic noise-free data for model parameters
using BA .................................................................................................. 102
Figure 5.7 Histograms of relative frequency of SP noise-free and noisy synthetic data
by M-H algorithm .................................................................................... 103
Figure 5.8 Error energy (a) and inversion result of multiple synthetic data for 600
iterations .................................................................................................. 104
Figure 5.9 Change of the model parameters (a-e) Source-1 and (f-j) Source-2 for
multiple synthetic data sets according to iteration number ..................... 105
Figure 5.10 Inversion result of the Bavarian field anomaly for 300 iteration (a) and
error energy according to the iteration number (b) with BA ................... 106
Figure 5.11 Convergence of the model parameters for Bavarian field anomaly
according to the iteration number (a-e) ................................................... 107
Figure 5.12 Inversion result of the Malachite field anomaly for 300 iteration (a) and
error energy according to the iteration number (b) with BA ................... 108
Figure 5.13 Convergence of the model parameters for Malachite field anomaly
according to the iteration number (a-e) ................................................... 108
Figure 5.14 Inversion result of the Süleymanköy field anomaly for 300 iteration (a)
and error energy according to the iteration number (b) with BA ............ 110
Figure 5.15 Convergence of the model parameters for Süleymanköy field anomaly
according to the iteration number (a-e) ................................................... 110
Figure 5.16 Inversion results of BA from the KTB Borehole (Germany) field SP data
set using vertical cylinder model (Case-1) for (a) the comparison of the
observed and calculated anomalies and (b) change of the error energy .. 112
Figure 5.17 The results of BA from the KTB Borehole (Germany) SP data set using
inclined sheet model (Case-2). (a) change of the error energy and (b) the
comparison of the observed and calculated anomalies ............................ 113
Figure 5.18 Subsurface structure for multiple inclined sheet model of the KTB
Borehole (Germany) according to the calculated anomalies by BA ....... 114
Figure 5.19 Inversion result of the Gravity noise-free data for 300 iteration (a) and
error energy according to the iteration (b) with Bat Algorithm .............. 119
Figure 5.20 Convergence of the model parameters for Gravity noise-free theoretical
data according to the iteration number (a-e) ............................................ 119

xii
Figure 5.21 Inversion result of the Gravity noisy data for 300 iteration (a) and error
energy according to the iteration (b) with Bat Algorithm ....................... 120
Figure 5.22 Convergence of the model parameters for Gravity noisy theoretical data
according to the iteration number (a-e) ................................................... 120
Figure 5.23 Error energy maps of gravity synthetic noise-free data for model
parameters using BA ............................................................................... 121
Figure 5.24 Histograms of relative frequency of Gravity noise-free and noisy synthetic
data by M-H algorithm. ........................................................................... 123
Figure 5.25 Inversion result of the Camaguey field anomaly for 300 iteration (a) and
error energy according to the iteration number (b) with BA. .................. 123
Figure 5.26 Convergence of the model parameters for Camaguey field anomaly
according to the iteration number (a-e) ................................................... 124
Figure 5.27 Histograms of relative frequency of Camaguey field data by M-H
algorithm.................................................................................................. 124
Figure 5.28 Inversion result of the Quebec field anomaly for 300 iteration (a) and error
energy according to the iteration number (b) with BA ............................ 125
Figure 5.29 Convergence of the model parameters for Quebec field anomaly
according to the iteration number (a-e) ................................................... 126
Figure 5.30 Histograms of relative frequency of Quebec field data by M-H algorithm
................................................................................................................. 126
Figure 6.1 Simplified flowchart of the PEA ............................................................ 130
Figure 6.2 Solutions of the sphere test function by PEA with 10 generation .......... 134
Figure 6.3 Solutions of the Rosenbrock’s test function by PEA with 10 generation
................................................................................................................. 135
Figure 6.4 Solutions of the Rastigin’s test function by PEA with 10 generation .... 136
Figure 6.5 Solutions of the Himmelblau’s test function by PEA with 10 generation..
................................................................................................................. 137
Figure 6.6 Solutions of the Beale test function by PEA with 10 generation ........... 138

xiii
LIST OF TABLES

Page

Table 1.1 Characteristics of the metaheuristic algorithms and derivative-based


methods........................................................................................................ 2
Table 3.1 Current applications of some metaheuristic algorithms for geophysical
inversions between 2016 to 2022 .............................................................. 50
Table 4.1 Parameter tuning study for CSA with the theoretical data without noise for
SP anomaly (The best result of the parameter tuning study is shown in
boldface type) ............................................................................................ 60
Table 4.2 Solutions based on the theoretical data sets of SP anomaly by CSA. ........ 62
Table 4.3 CSA estimations, mean values and standard deviations (SD) for the synthetic
data obtained from M-H algorithm............................................................ 67
Table 4.4 True values, parameter bounds and CSA estimations for the multiple
synthetic data ............................................................................................. 69
Table 4.5 The solution by CSA in comparison with the prior studies (The Bavarian
woods anomaly)......................................................................................... 70
Table 4.6 The solution by CSA in comparison with the prior studies (The Süleymanköy
anomaly) .................................................................................................... 71
Table 4.7 The solution by CSA in comparison with the prior studies (The Malachite
mine anomaly) ........................................................................................... 72
Table 4.8 The inversion results by CSA (KTB Borehole Anomaly) using vertical
cylinder SP model (Case-1) ....................................................................... 73
Table 4.9 The solution by CSA for Case-1 in comparison with the prior studies (The
KTB Borehole anomaly) ........................................................................... 74
Table 4.10 The inversion results by CSA (KTB Borehole Anomaly) using inclined
sheet SP model (Case-2) ............................................................................ 75
Table 4.11 The solution by CSA for Case-2 in comparison with the prior studies (The
KTB Borehole anomaly) ........................................................................... 76
Table 4.12 CSA estimations together with mean values and standard deviations (SD)
for field data sets obtained from M-H algorithm....................................... 78

xiv
Table 4.13 Parameter tuning study for CSA with the theoretical data without noise for
gravity anomaly (The best result of the parameter tuning study is shown in
boldface type) ............................................................................................ 81
Table 4.14 Solutions based on the theoretical data sets of gravity anomaly by CSA..
................................................................................................................... 82
Table 4.15 CSA estimation, mean and standard deviation values of noise-free and
noisy theoretical data ................................................................................. 86
Table 4.16 Parameter estimation results of CSA and search space boundaries of
Camaguey field data .................................................................................. 87
Table 4.17 Parameter estimation results of CSA and search space boundaries of
Quebec field data ....................................................................................... 89
Table 4.18 CSA estimation, mean and standard deviation values of Camaguey and
Quebec field data ....................................................................................... 91
Table 5.1 Parameter tuning study for BA with the theoretical data without noise for SP
anomaly (The best result of the parameter tuning study is shown in boldface
type) ........................................................................................................... 97
Table 5.2 Solutions based on the theoretical data sets of SP anomaly by Bat Algorithm
................................................................................................................... 99
Table 5.3 Bat Algorithm estimations for SP model parameters, mean and standard
deviation values of noise-free and noisy theoretical data ........................ 103
Table 5.4 True values, parameter bounds and BA estimations for the multiple synthetic
data........................................................................................................... 104
Table 5.5 The solution by BA in comparison with the prior studies (The Bavarian
woods anomaly)....................................................................................... 107
Table 5.6 The solution by BA in comparison with the prior studies (The Malachite
mine anomaly). ........................................................................................ 109
Table 5.7 The solution by BA in comparison with the prior studies (The Süleymanköy
anomaly) .................................................................................................. 111
Table 5.8 The results by BA (KTB Borehole Anomaly) using vertical cylinder SP
model (Case-1) ........................................................................................ 111
Table 5.9 The solution by BA for Case-1 in comparison with the prior studies (The
KTB Borehole anomaly) ......................................................................... 112

xv
Table 5.10 The inversion results by BA (KTB Borehole Anomaly) using inclined sheet
SP model (Case-2) ................................................................................... 113
Table 5.11 The solution by BA for Case-2 in comparison with the prior studies (The
KTB Borehole anomaly) ......................................................................... 113
Table 5.12 Parameter tuning study for BA with the theoretical data without noise for
gravity anomaly (The best result of the parameter tuning study is shown in
boldface type). ......................................................................................... 116
Table 5.13 Solutions based on the theoretical data sets of Gravity anomaly by Bat
Algorithm................................................................................................. 118
Table 5.14 Bat Algorithm estimations for Gravity model parameters, mean and
standard deviation values of noise-free and noisy theoretical data ......... 122
Table 5.15 The solution by BA in comparison with the prior studies (The Camaguey
anomaly) .................................................................................................. 125
Table 5.16 The solution by BA in comparison with the prior studies (The Quebec
anomaly) .................................................................................................. 127
Table 6.1 Control parameters and explanations of the PEA .................................... 130
Table 6.2 Details of main test functions. ................................................................. 133

xvi
CHAPTER ONE
INTRODUCTION

Geophysics is a science that deals with the internal properties of the Earth by
applying the principles of physics. Physical properties due to compositional
heterogeneity are evaluated to determine the potential of Earth's environmental and
energy resources (Chauhan, 2017). To this end, geophysical surveys can be conducted
using either passive methods, in which a natural field such as the Earth's magnetic or
gravitational field is detected and measured, or active methods, which aim to detect
the response of the subsurface to a signal sent into the ground (Parasnis, 1986). In
geophysics, observed or measured data generated by either natural or artificial sources
are used to describe the internal properties of the Earth. This attempt leads to an inverse
problem in geophysics (Zhdanov, 2015). Such an inverse problem can be solved by
linearized (i.e., least squares) or nonlinear techniques to find the solution to
geophysical problems. Therefore, inversion studies play an important role in
estimating model parameters from geophysical anomalies. Derivative-based
algorithms such as least squares are widely used for this purpose. Recently, nature-
inspired metaheuristic algorithms have been preferred for parameter estimation.
Metaheuristics use some search algorithms based on inspirations from processes in
nature, behaviors of animals, etc. to find a solution to the problem at hand. The aim of
this thesis is to investigate the performance of three newly introduced metaheuristic
algorithms (cuckoo search algorithm, bat algorithm, and perturbation-based
evolutionary algorithm) for the inversion of geophysical anomalies.

1.1 Motivation of Thesis

In the science and technology applications, the oldest and most popular
metaheuristic methods can be expressed as genetic algorithm-GA (Holland, 1975),
simulated annealing-SA (Kirkpatrick et al., 1983), particle swarm optimization-PSO
(Kennedy and Eberhart, 1995), ant colony algorithm-ACA (Dorigo, 1992), and
differential evolution-DE (Storn and Price, 1995, 1997; Storn, 1996). On the other
hand, growing technology offers new nature or bio inspired metaheuristic algorithms

1
for the geophysical engineering problems. Therefore, this thesis is motivated by the
lack of the applications of newer metaheuristic algorithms into the geophysical inverse
problems.

The derivative-based optimization algorithms use deterministic approaches, and


they need a well-designed initial model to reach a solution. It converges to the solution
faster than a metaheuristic method, but it can end up with a local minimum instead of
the global one.

On the other hand, the metaheuristic methods are independent of the initial model,
and these methods can scan a large search space effectively to reach a global minimum.
They are also able to avoid the local minima. Due to the characteristics mentioned
above, the metaheuristic methods are frequently preferred despite their higher
computational cost caused by a large number of forward solutions (Göktürkler, 2011).
Metaheuristic methods can be applied to any engineering problems. Table 1.1
summarizes pros and cons of metaheuristic algorithms and derivative-based methods.

Table 1.1 Characteristics of the metaheuristic algorithms and derivative-based methods

Metaheuristic Algorithms Derivative-based Methods


• Successful in reaching global optima • Possibility of reaching local optima
and avoiding local optima instead of global optima
• Not depending on an initial model • Depending on a good initial model
• Not required derivative-based • Required derivative-based
calculations calculations
• Large scale searching • Limited scale searching
• High computational cost depending • Low computational cost because of
on objective functions and wide good initial model and narrow range
range of search space of search space

Additionally, many geophysical applications with metaheuristic algorithms shows


that they are robust and accurate solutions can be yielded with metaheuristic methods
(Toklu, 2014). Therefore, metaheuristic methods offer successful and efficient results

2
which are close to the true results, for the estimations of geophysical parameters of
various geophysical inversions without a well-constructed starting model requirement
(Balkaya et al., 2017).

This thesis aimed to investigate three new nature inspired metaheuristic methods in
geophysical applications as Cuckoo Search Algorithm (CSA), Bat Algorithm (BA) and
Perturbation-based Evolutionary Algorithm (PEA) for the inversion of gravity and
self-potential synthetic and field data. The discussions of advantages and
disadvantages of proposed metaheuristic methods will develop the literature of
geophysics because of these metaheuristic methods have not been used for the
inversion of the geophysical data yet.

1.2 Thesis Objectives

The objectives of this thesis were determined by taking into account the current
developments and gaps in the literature. Both synthetic and field data sets from self-
potential and gravity methods were used in the test studies including parameter
estimation and uncertainty appraisal analysis. Self-potential anomalies are occurred by
natural electrical fields of the Earth. On the other hand, gravity anomalies are occurred
by changing of the gravitational fields of the Earth. Self-potential and gravity
anomalies, that used in thesis, are mathematically modelled with for various
geometrical shapes for the inversion and uncertainty appraisal analysis. The objectives
of this thesis are listed below for self-potential and gravity data sets.

• Arrangement of Cuckoo Search Algorithm (CSA) with Lévy flights instead of


conventional random walks and Bat Algorithm (BA) for geophysical inversion
problems.
• Mathematical modeling of self-potential anomaly for infinitely long horizontal
cylinder to generate without noise synthetic data for the inversions with CSA
and BA algorithm.

3
• Investigation of parameter tuning study to determine best values of algorithm-
based control parameters for CSA and BA using the noise-free synthetic self-
potential data.
• Inversion studies with CSA and BA for noise-free and noisy (noise-
contaminated with pseudo random numbers) synthetic self-potential data sets
to calculate model parameters and compare them with their true values.
Additionally, calculation of root mean square (rms) errors of without noise and
noise-contaminated synthetic data sets according to the error energy and
presentation of observed and calculated anomaly results.
• Changing of the model parameters by iteration number for without noise and
noise-contaminated synthetic self-potential anomalies using CSA and BA.
• Investigation of uncertainty solutions obtained from CSA and BA via
distribution histograms of relative frequency by Metropolis-Hasting (M-H)
sampling algorithm for the model parameters of without noise and noise-
contaminated synthetic self-potential anomalies.
• Inversion studies for self-potential field data sets from Bavarian woods in
Germany, Süleymanköy in Türkiye, and Malachite mine in Colorado (USA)
using CSA and BA.
• Comparison of the inversion results of the estimated model parameters and
rms values of self-potential field data sets using CSA and BA with the results
from the published literature.
• Investigation of uncertainty solutions obtained from CSA and BA via
distribution histograms of relative frequency by Metropolis-Hasting (M-H)
sampling algorithm for the model parameters of field self-potential anomalies.
• Mathematical modeling of gravity anomaly for infinitely long horizontal
cylinder to generate noise free synthetic data for the inversions with CSA and
BA algorithm.
• Investigation of parameter tuning study to determine best values of algorithm-
based control parameters for CSA and BA using the noise-free synthetic
gravity data.
• Inversion studies with CSA and BA for noise-free and noise-contaminated
synthetic gravity data sets to calculate model parameters and compare them

4
with their true values. Additionally, calculation of rms errors of without noise
and noise-contaminated synthetic data sets according to the error energy and
presentation of observed and calculated anomaly results.
• Changing of the model parameters by iteration number for without noise and
noise-contaminated synthetic gravity anomalies using CSA and BA.
• Investigation of uncertainty solutions obtained from CSA and BA via
distribution histograms of relative frequency by Metropolis-Hasting (M-H)
sampling algorithm for the model parameters of without noise and noise-
contaminated synthetic gravity anomalies.
• Inversion studies for gravity field data sets from Camaguey in Cuba and
Quebec in Canada using CSA and BA.
• Comparison of the inversion results of the estimated model parameters and
rms values of gravity field data sets using CSA and BA with the results from
the published literature.
• Investigation of uncertainty solutions obtained from CSA and BA via
distribution histograms of relative frequency by Metropolis-Hasting (M-H)
sampling algorithm for the model parameters of field gravity anomalies.
• Introducing of Perturbation-Based Evolutionary Algorithm (PEA) as a new
metaheuristic optimization algorithm for optimization problems.
• Testing of efficiency of the PEA with various test functions.

1.3 Structure of Thesis

This thesis is organized as seven chapters according to the objective of the thesis.
Thesis starts with Chapter One, which is titled “Introduction” in the thesis, and this
section summarizes the fundamentals of the geophysics and geophysical inversions.
Motivations, objectives, and structure of the thesis are also explained in the first
chapter for better understanding of the purpose of the study and methodology. Chapter
Two explains theory and applications of popular metaheuristic methods for the
inversions of geophysical anomalies. General definitions, historical perspectives,
solution methods and detailed explanations of algorithms is given for each
metaheuristic algorithms. Chapter Three focuses on detailed literature review of

5
metaheuristic algorithms used in geophysical inversion studies. Literature review
includes the investigations of the applications of four popular and several relatively
new metaheuristic algorithms on various geophysical anomalies. Chapter Four gives
general explanations of Cuckoo Search Algorithm and performs inversion studies with
noise-free and noisy synthetic data sets and field data sets for self-potential and gravity
anomalies with uncertainty solutions via distribution histograms of relative frequency
by Metropolis-Hasting (M-H) sampling. Chapter Five gives general explanations of
Bat Algorithm and performs inversion studies with noise free and contaminated with
pseudo random number synthetic data sets and field data sets for self-potential and
gravity anomalies. Additionally, uncertainty solutions via distribution histograms of
relative frequency by Metropolis-Hasting (M-H) sampling are presented. Chapter Six
proposes a new metaheuristic algorithm named Perturbation-based Evolution
Algorithm and its performance with various test functions. Chapter Seven summarizes
the main results of the inversions using Cuckoo Search Algorithm, Bat Algorithm and
Perturbation-based Evolution Algorithm. Advantages and disadvantages of the
algorithms are also discussed compared to other metaheuristic algorithms.

6
CHAPTER TWO
AN OVERVIEW OF METAHEURISTIC METHODS IN GEOPHYSICS

2.1 Introduction to Metaheuristics

Meta means ‘beyond’ or ‘higher level’ and heuristic means “find”, “discover” and
“search” in Greek. Metaheuristic algorithms are generally described as problem
independent algorithms because they can easily be adapted to a variety of the problems
(Peres and Castelli, 2021). There is no a general agreement in the literature for the
exact definitions for the terms heuristics and metaheuristics. Usually, they are used
interchangeably. Recently, metaheuristic has become a term including stochastic and
global exploration methods (Gandomi et al., 2013).

Because of the certain limitations of the classical and heuristic methods,


metaheuristic methods were developed to eliminate these disadvantages caused by
these limitations. The main advantage of metaheuristics that any specific knowledge
does not require about the optimization problem for the implementation of desired
metaheuristic algorithm. Therefore, metaheuristics can be used a wide range of
optimization problems. However, the metaheuristics should be designed for each
individual optimization problem separately. Metaheuristic methods find the solutions
iteratively according to their predefined stochastic rules. Thus, escaping from local
optimum, balancing exploration and exploitation and independent searching according
to the initial parameters can be achieved (Krüger, 2017).

Metaheuristic methods offer efficient alternative to get acceptable solutions with


reasonable time consumption for a complex problem based on trial and error. if the
complexity of the optimization problem is extremely high, it may be very difficult or
even impossible to search every possible solution. In that case, a feasible solution is
aimed to find in an acceptable time. The feasible solution may not be the best solution
in every time and metaheuristic methods do not guarantee to find the best solutions
even if the algorithm will work faultless (Yang, 2014; Yang et al., 2014). The main
purpose of the metaheuristic method is to obtain good quality solutions, and this can

7
be determined by optimality of the solution that is how close to the optimal solution
(Yang, 2011).

Any metaheuristic algorithm has two main components including intensification


also called exploitation and diversification also called exploration. Diversification
includes generating of diverse solutions for the exploration of the search space on the
global scale. On the other hand, intensification focuses on the search in a local area
using the information of a good solution is found in this area. This combination is used
for the selection of the best solutions. On the other hand, escaping from the local
optima is possible with increasing the diversity of the solutions by the diversification
with randomization. The good combination of intensification and diversification can
provide the achievable global solution (Blum and Roli 2003). For example, low
diversification and high intensification may cause the system trapped in local optima.
After being trapped in local optima, it is difficult or maybe impossible to escape from
there and find the global optimum. On the other hand, high diversification with low
intensification may cause difficult convergence for the system. Therefore, arrangement
of the diversification and intensification levels is a major optimization problem for
metaheuristic methods (Yang, 2011).

Metaheuristic methods use different ways for the balance between diversification
and intensification. The combination of certain randomization and a deterministic
procedure can be used to achieve diversification and intensification. Additionally,
validation and testing of any metaheuristic algorithm are crucial to generate feasible
solutions in the search space (Gandomi et al, 2013).

General characteristics of metaheuristic methods can be summarized as follows:

• Search for global solution is aimed by metaheuristic methods,


• Main purpose is exploration of the search space efficiently for optimal
solution,
• Metaheuristic methods include wide range of techniques from simple local
search to complex learning.

8
• Approximate solutions are found iteratively with metaheuristic methods,
• Metaheuristic methods may have specific approaches for avoiding local
optima in search space,
• Metaheuristic methods can implement any optimization problem,

According to the general characteristics and operation methods of metaheuristic


methods, it can be said that high level strategies are used in various types of
metaheuristic algorithms for search space to find an optimal solution (Blum and Roli,
2003). Searching strategies of various metaheuristic algorithms are based on types of
metaheuristics and inspired mechanisms. Therefore, classifications of metaheuristic
methods are important to understand the mechanism of the algorithms.

2.2 Classification of Metaheuristics

Metaheuristic methods are described and classified in various ways depending on


the characteristics of their algorithms. According to the origins, inspired mechanisms
and randomization methods, several classifications can be defined for metaheuristic
algorithms (Blum and Roli, 2003). Classifying of metaheuristics are summarized for
various evaluation criteria to understand their developments and implementations to
any optimization problem.

According to the origins of the metaheuristic methods, general behavior of the


metaheuristic algorithm is based on rules from nature or mathematical principles.
Many systems and process can be found for the inspiration of a metaheuristic
algorithm in nature such as evolution, colony, or swarm behavior, annealing process
and immune systems. Metaheuristics, which are designed with these phenomena, are
called as nature inspired metaheuristic methods in the literature. Growing demand of
nonlinear and high dimensional optimization problems lead increasing of nature
inspired algorithms depending on high computing power and memory storage capacity
with technological evolution. On the other hand, nature inspired algorithms from
biological environment depend on the adaptation of the system to changeable
biological environments. Biological system inspirations include evolution, neural

9
networks, ant colony, bee colony, particle swarm, immune systems, and bacteria
systems. Nature inspired metaheuristic algorithms are used for various optimization
applications to obtain better and flexible solutions than classical techniques.
Mathematically founded metaheuristic methods such as Tabu Search Algorithm,
which is based on mathematical principles, include local search principles of solution
space, neighborhoods for solutions and definitions for distance between solutions
(Krüger, 2017).

Metaheuristic methods can be classified in the literature based on the number of


solutions as single solution methods and population-based methods. In the literature,
Tabu Search, Iterated Local Search and Simulated Annealing are developed as a single
solution method, which can be called as trajectory-based methods. Two stage solution
approach including generation and replacement is used in the single solution method
to find the single solution. First, candidate solutions are produced using current
solution in generation stage. Then, suitable solution is chosen from produced solutions
as new current solution in replacement stage. These stages continue until desired
solution is obtained according to the quality of the solution. In single solution method,
initial solution can be found randomly or using a heuristic approach (Gogna and Tayal,
2013). On the other hand, population-based metaheuristic methods, which are mostly
nature inspired metaheuristics, offer new solution as a combination of existing
solutions. Popular population-based metaheuristic methods used for the optimization
problems in engineering literature can be given as Genetic Algorithms, Particle Swarm
Optimization, Bee Colony Optimization, Ant Colony Optimization and Artificial
Immune Systems (Krüger, 2017). Population based methods use generation and
replacement approach to spread a group of solutions over the search space. Population
based methods use various ways to perform generation, selection, and organization of
searching memory based on inspired system of the algorithms. Efficient algorithm
needs to a well determined initial populations. On the other hand, number of iterations
and minimum value of objective function are other important parameters about
stopping criteria for the efficiency of the algorithm (Gogna and Tayal, 2013).

10
Metaheuristic methods are also classified depending on the decision process.
According to the types of rules of the decision process, metaheuristic methods can be
defined as deterministic and stochastic. Deterministic set of rules are used for the
solving an optimization problem in deterministic methods. Tabu Search algorithm can
be given as deterministic method. If the decision process is based on randomization,
metaheuristic methods are classified as a stochastic algorithm. It is possible to say that
most of current metaheuristic methods used in the literature are stochastic because of
the randomization process (Krüger, 2017).

2.3 Genetic Algorithm

GA was defined by Louis and Li, (2000) as a search algorithm that uses natural
genetic methods, to find a solution not obtained by traditional calculation methods in
large and non-linear search space. Therefore, GA is called a solution technique to find
the best solution or solutions of an optimization problem based on the natural evolution
mechanism (Goldberg, 1989a). According to the evolution approach, the logic of
natural selection as survival of population that adapt to the natural environment and
elimination of those that cannot adapt is considered by GA to search for the best
solution using these rules (Holland, 1975).

Pre-defined number of individuals in the initial population are determined randomly


for the solution of the optimization problem with GA. After the determination of the
random initial population, genetic and evolutionary processes must be determined.
Genetic process includes crossover and mutation. On the other hand, evolutionary
process includes selection. Each individual in the solution set is called a chromosome.
Chromosomes are occurred by series of symbols named genes (Goldberg, 1989b).

Basic steps of the operation strategy of the GA can be summarized as follows for
the solution of an engineering optimization problem.

• Step 1: A solution set which is named “population”, is created depending on


possible solution codes named “chromosomes”. Firstly, the population

11
number is determined. There is not any certain number for the initial number
of population. However, initial population can be taken between 100 and 300.
Then, the population number is redetermined randomly with the iterations of
the algorithm.
• Step 2: The quality of each chromosome is evaluated by the fitness function.
Determination of the fitness function is depended on selection criteria of the
optimization problem in GA such as minimization or maximization and the
success of the GA algorithm is mostly depended on the efficiency and
sensitiveness of the fitness function.
• Step 3: A new population called “generation” is created by performing
genetic operators such as crossover, mutation, and selection according to the
fitness value of the chromosomes. Various methods can be used for the
selection in GA such as Roulette Wheel Selection, Rank Selection, Steady
State Selection, Tournament Selection, Elitism Selection and Boltzmann
Selection.
• Step 4: Old chromosomes are removed, and the new ones are replaced.
Constant number of the population is provided with the new chromosomes.
• Step 5: Fitness values of all chromosomes are recalculated.
• Step 6: Algorithm runs repeatedly for many numbers of populations. Best
chromosome is accepted as a solution because best individuals are kept in the
population.

According to the design steps of the GA for an optimization problem, pseudo code
of GA can be given in Fig. 2.1 depending on objective function of f(x), fitness of F,
crossover, and mutation probability of pc and pm for the iteration number t of the
algorithm.

Genetic operators define the processes applied on the existed population. These
processes aim to produce better new generations and to expand the search space of the
GA. Three standard genetic operators are used in GA as reproduction, crossover, and
mutation.

12
Figure 2.1 Pseudo code of GA (Modified from Yang, 2014)

Genetic operators define the processes applied on the existed population. These
processes aim to produce better new generations and to expand the search space of the
GA. Three standard genetic operators are used in GA as reproduction, crossover, and
mutation.

Genetic algorithms are a nature inspired optimization method that randomly finds
the best solution in the loop. The fitness function, which determines the best solution,
chooses among new solutions and the crossover operator is used to produce a better
result. The solutions can be trapped into a local solution in this process. Mutation
operator is used to jump the local optima and to reach global optima at the end of the
loop of the algorithm (Haupt and Haupt, 2004).

There are many important differences between GA and traditional optimization


methods according to the characteristics of the GA as a nature inspired metaheuristic
method. GA uses the coded versions of the optimization parameters instead of the real
parameters. GA uses fitness function values instead of derivative values. GA does not
based on deterministic rules. Searching process starts from multiple points in GA
instead of one point. Therefore, possibility of getting trapped in local optima is
extremely low. However, traditional methods stop searching when a local optimum
was found. If an optimization problem has more than one optimum points as solution,

13
traditional models are not useful to find the global optimum points. GA finds all local
optima and compare the solutions to determine the global optimum point.
Additionally, GA is useful for the complex optimization problems that have multiple
parameters (Zuo, 1995).

2.4 Particle Swarm Optimization

Particle swarm optimization (PSO) algorithm was proposed by Kennedy and


Eberhart (1995), as a swarm-based nature inspired optimization technique. This
algorithm focuses to simulate the social behavior of many animals such as insects,
herds, birds, and fish. According to the behavior of the animals in a swarm, an animal
cooperates with other individuals, and it adapts to find food depending on changing its
search habits according to the learning experiences. This social interaction can be
modeled for the optimization problems with PSO. In PSO, the population is called a
swarm and each bird of the swarm is called a particle.

The main design idea of the PSO algorithm is related to two studies of evolutionary
algorithm and artificial life. PSO uses a swarm mode related to the basics of
evolutionary algorithm to simultaneously search a large solution space to optimize the
objective function. On the other hand, the artificial life of the swarm examines the
behavior of social animals with the artificial life theory. To construct the swarm
artificial life systems with cooperative behaviors of individuals in the swarm, five
basic principles were proposed as follows (Van Den Bergh, 2001):

• Proximity: The swarm should include simple space and time calculations.
• Quality: The swarm must be able to detect and respond to quality change in the
environment.
• Diverse response: The swarm should not limit the way to obtain resources in a
narrow scope.
• Stability: The behavior mode of the swarm should stable in every
environmental change.

14
• Adaptability: The behavior mode of the swarm should change if every
environmental change is worth it.

The positions and velocities of the particles can be updated depending on changing
the environment in PSO for proximity and quality requirements of the optimal solution
in the solution space. Particles can maintain their stable motion in search space despite
changing their type of motion for adaptation to the environmental changes. Thus, the
five principles can be carried out in PSO (Wang et al., 2018).

PSO is preferred for solution of the optimization problems because of its simplicity
and rapid convergence characteristics among other population-based methods. Each
particle is a candidate solution in PSO algorithm. These particles aim to find the most
suitable solution by search space. The basic logic of the algorithm can be explained as
reporting to the bird population (swarm) for food source and searching best food
sources by birds (particles). In this way, almost the entire search space can be searched
by birds and a solution, which is close to the best solution, can be found. Birds in the
swarm have their own memories, and they also know where the best food source is.

Basic steps of the operation strategy of the PSO can be summarized as follows for
the solution of an engineering optimization problem (Ekinci et al., 2020).

• Step 1: The PSO algorithm starts with defining the objective function and
setting the population. The initial values of location and velocity for each
particle are assigned randomly.
• Step 2: The fitness value of each particle is calculated according to the given
objective function.
• Step 3: Each individual's own best values (pbest) and the best of all individuals
(gbest) are obtained. Calculated fitness value in the previous step is compared
with the best value of pbest. According to the comparison, if the first result is
better than the current best pbest, the new result is defined as new pbest.
• Step 4: Best global particle is found. Fitness values of each particle calculated
in Step 2 is compared with best global optimum solution (gbest). If there is a

15
better result than current gbest, the result is defined new gbest. Comparison
process is carried out for all particles.
• Step 5: New velocity and location values are calculated. The equations of new
velocity (𝑣𝑖𝑡+1 ) and location (𝑥𝑖𝑡+1 ) are presented as follow:

𝑣𝑖𝑡+1 = 𝑣𝑖𝑡 + 𝛼𝜖1 [𝑔∗ − 𝑥𝑖𝑡 ] + 𝛽𝜖2 [𝑥𝑖∗(𝑡) − 𝑥𝑖𝑡 ] (2.1)

𝑥𝑖𝑡+1 = 𝑥𝑖𝑡 + 𝑣𝑖𝑡+1 (2.2)

where, 𝑥𝑖∗(𝑡) is the current best for particle i, g* is the current global best at
current iteration t, α and β are the learning parameters, 𝜖1 and 𝜖2 are two random
vectors that takes the values between 0 and 1.
• Step 6: The processes between Step 2 and Step 7 are repeated until the achieved
conditions of stopping. There are two crucial criteria for the consideration of
the stopping criteria. As a first criteria, stopping condition should not cause
premature convergence because of trapping in the local best point. Secondly,
if the stopping condition needs high computation for the calculation of the
fitness function, the computation cost increases. Stopping criteria can be
determined according to the predetermined maximum number of iterations,
achieved result and improvement of the process in certain period of time.

According to the design steps of the PSO for an optimization problem, pseudo code
of PSO can be given in Fig. 2.2 depending on objective function of f(x), fitness of F,
crossover, and mutation probability of pc and pm for the iteration number t of the
algorithm.

16
Figure 2.2 Pseudo code of PSO (Modified from Yang, 2014)

The PSO algorithm has many advantages as a nature inspired metaheuristic


algorithm for the optimization problems. PSO is simple to implement with few
parameters to adjust. PSO is robust and it can be run with parallel computation. In
optimization problems, PSO has high probability and efficiency to find the global
optimum. The convergence of the PSO is very fast for the computation. Hence, PSO
has low computational cost. PSO does not overlap or mutate during all these
convergence processes. For these reasons, PSO can be effective for solving problems
that can be difficult to find with the mathematical models. On the other hand, the
disadvantages of the PSO algorithm can be expressed as; definition of the design
parameters is difficult, scattering problems cannot be solved and early converge or
stuck at the local minimum in complex problems can be possible in PSO (Abdmouleh
et al., 2017).

2.5 Differential Evolution Algorithm

Differential Evolution Algorithm (DEA) is a population-based metaheuristic


optimization technique based on genetic algorithm. DEA was developed by Storn and
Price (1995) for effective results in optimization problems with good convergence
properties (Storn and Price, 1995; Storn and Price, 1997). DEA investigates many
points at the same time and better results are searched for the solution of the problem

17
with the operators during iterations. Variables are defined with their real values in
DEA, unlike classical binary in GA. Encoding and decoding does not need in DEA
(Yang, 2014).

On the other hand, Storn and Price (1995) have worked on increasing the solution
performance of problems with some changes in genetic operators by coding with real
values. DEA is use crossover, mutation and selection operators which are also used in
GA. However, each operator does not apply sequentially to the entire population in
DEA. Chromosomes are evaluated individually, and a new individual is obtained by
using the other three randomly selected chromosomes using mutation and crossover
operators. The fitness values of the new chromosome, which was obtained with the
existing chromosome, is compared and the new individual with better fitness value is
transferred to the next population as a new individual (Das and Suganthan, 2010).

Evolutionary algorithms operate based on mutation, reproduction, evaluation, and


selection operators. Apart from the GA, which operates based on the previously
defined probability distribution function, the DEA uses a mutation operator based on
the difference of randomly selected vectors. The simple mutation operation used in the
DEA has some benefits for the optimization problems. It improves the performance of
the algorithm, enables the exploration of new search spaces, and causes stable solution.
Additionally, the DEA has some advantages about fast convergence, simplicity, easy
usability, ability to converge to the global optimum, parallelism, low computational
cost, easy adaptation to discrete and complex optimizations, not requiring derivative
operations, and simple use for time-dependent and noisy objective functions (Mayer
et al., 2005).

Basic steps of the operation strategy of the DEA can be summarized as follows for
the solution of an engineering optimization problem (Karaboğa and Ökdem, 2004).

• Step 1: Initial population of the DEA is determined. The variable numbers of


the problem determine each size (number of genes) of the chromosome. The
chromosome numbers, which are determined by the user, should be taken as

18
higher than three. Because, to produce new chromosomes, three chromosomes
are required in DEA.
• Step 2: Differential weight and crossover probability of the solution in the
initial population are calculated.
• Step 3: Three chromosomes that are different from each other and apart from
the chromosome in mutation process are randomly selected. The difference of
the first two of these chromosomes (𝑥𝑞𝑡 -𝑥𝑟𝑡 ), that selected randomly, is
calculated. Then, calculated difference is multiplied by the F parameter of the
chromosome. F (often referred to as the differential weight) parameter usually
takes values between 0 and 2. The obtained difference chromosome is summed
with the selected third chromosome (𝑥𝑝𝑡 ). Therefore, the chromosome for
crossover process is obtained because of mutation as shown in the equation
below.

𝑣𝑖𝑡+1 = 𝑥𝑝𝑡 + 𝐹(𝑥𝑞𝑡 − 𝑥𝑟𝑡 ) (2.3)

• Step 4: Candidate solution is generated according to the crossover rate in


crossover process. This crossover method used in DEA includes addition of
crossover rate to the uniform crossover used in binary GA. In uniform
crossover, each gene is evaluated separately and selected from one of the two
parent chromosomes with equal probability. However, crossover probability is
used in DEA instead of an equal probability. For each parameter of vj,i fitness
value is shown in Eq. 2.4 according to the crossover rate.

𝑡+1
𝑡+1
𝑣𝑗,𝑖 𝑖𝑓 𝑟𝑖 ≤ 𝐶𝑟 𝑜𝑟 𝑗 = 𝐽𝑟
𝑢𝑗,𝑖 ={ 𝑡 (2.4)
𝑥𝑗,𝑖 𝑖𝑓 𝑟𝑖 > 𝐶𝑟 𝑎𝑛𝑑 𝑗 ≠ 𝐽𝑟

• Step 5: Evaluation of the current generation and the newly produced


chromosomes with the selection operator creates a new generation. The
probability of chromosomes in the next generation depends on their fitness.
There is no need for selection operators with complex procedure for selection
because the comparison is performed individually in DEA. The chromosome
with the highest fitness among the compared chromosomes is assigned as the

19
individual of the new generation. For the parameter in t+1 generation,
comparison of fitness value (𝑢𝑖𝑡+1 ) and objective value (𝑓(𝑥𝑖𝑡 )) yields the
selection of the algorithm as shown in below.

𝑢𝑖𝑡+1 𝑖𝑓 𝑓(𝑢𝑖𝑡+1 ) ≤ 𝑓(𝑥𝑖𝑡 )


𝑥𝑖𝑡+1 ={ 𝑡 (2.5)
𝑥𝑖 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

• Step 6: According to the acceptance criteria, solutions are accepted or updated.


The aim for the acceptance is to obtain chromosomes with the best fitness value
and find the optimum. If stopping criteria is not provided, go back to the Step
2.

According to the design steps of the DEA for an optimization problem, pseudo code
of DEA can be given in Fig. 2.3 depending on weight F, crossover probability Cr,
random index Jr, randomly distributed number ri, fitness value ui for the parameter xi
in the any time or generation of t for the algorithm.

Figure 2.3 Pseudo code of DEA (Modified from Yang, 2014)

The DEA algorithm is a useful nature inspired metaheuristic algorithm for the
optimization of the objective function which can be either minimized or maximized.
DEA offers accurate and robust solution for the optimization problem including linear

20
or non-linear, differentiable, or non-differentiable objective functions. Hence, DEA
paid attention as a global optimization method with the characteristics of reliability,
easy to use and simple to implementation. In engineering design problems and real-
life applications, DEA yields efficient solutions with its self-organizing capability
(Vasuki, 2020).

2.6 Simulated Annealing

Simulated Annealing (SA) algorithm that is one of the most popular techniques for
optimization problems in engineering, was proposed by Kirkpatrick et al., (1983) as a
metaheuristic algorithm. The purpose of the algorithm is to obtain global minimum or
maximum value of any function or measurement for especially in optimization
applications of combinatorial problems that cannot be represented by mathematical
models (Kirkpatrick et al., 1983).

The basic ideas of the SA algorithm were first conducted by Metropolis et al.,
(1953). The SA algorithm is inspired by the cooling of materials in hot baths (known
as the annealing process). If the solid material is heated up to its melting point and
cooled again until it becomes to solid phase, the structural properties of the chilled
version of the solid material will depend on the cooling rate. For example, crystals of
the materials can increase with very slow cooling. However, rapid cooling can cause
defects in the crystal structure. According to the heating and cooling processes, the
annealing process can be simulated as a material-based system. Hence, Metropolis et
al., (1953) simulated the changes in energy of a cooling system until it reaches a
steady-state condition. Thirty years later, Kirkpartrick et al., (1983) suggested that this
simulation could be used to find solutions of the optimization problems which were
converged to the optimum.

The SA approach can be expressed as a variation of the local (neighborhood) search


technique, which is an iterative search technique from existing solutions to
neighboring solutions to find global optimum point. The local search technique for a
minimization problem employs a cooling strategy that the search is always improved.

21
However, this strategy converges to a local point rather than a global solution. A
satisfactory solution can be found by applying the algorithm with several different
initial solutions and by increasing the complexity of the neighborhoods with the
expansion of the objectives.

The solutions obtained from the search strategies for SA depend on the initial
solutions. However, a reliable metaheuristic approach in SA should depend on as little
as possible to the initial solution. In the cooling process, the search is performed to the
lowest point of the process which is included the initial solution. There can be some
uphill searching to the peaks of the process in the solution space. Because of the final
purpose is convergence to a minimum point, this must be done under control. In the
SA, uphill searching movements are allowed, but their frequency depends on a
probability function that changes with the progress of the algorithm (Metropolis et al.,
1953).

Basic steps of the operation strategy of the SA can be summarized as follows for
the solution of an engineering optimization problem (Göktürkler, 2011; Yang, 2014).

• Step 1: Determining the starting point and initial value for the model parameter.
Then, calculation of the final temperature based on the error energy (E)
according to the number of iteration number (N), observed temperature (T0)
and calculated temperature (Tc).

1 0 𝑐 2
𝐸 = ∑𝑁
𝑖=1(𝑇𝑖 − 𝑇𝑖 ) (2.6)
𝑁

• Step 2: Choosing a new set of parameters randomly in the parameter space and
calculating the error energy for the new parameters.

• Step 3: Calculating ΔE and observing the results:

a. If ΔE<0, accept the point

b. If ΔE≥0, accept according to the probability. If the random number is


uniformly distributed between 0 and 1, accept. If not, it is rejected.

22
• Step 4: Repeating the steps 2-3 to get the best parameter set with minimum
error.

According to the design steps of the SA for an optimization problem, pseudo code
of SA can be given in Fig. 2.4 depending on initial temperature T0, final temperature
Tf, maximum number of iteration N, random number r, for the best parameter x* and
global minimum f* of the algorithm.

Figure 2.4 Pseudo code of SA (Modified from Yang, 2014)

SA is a common and robust algorithm that can handle with high order nonlinear
problems, noisy data, and the optimizations with many constraints. Its flexibility and
ability to find a global best solution can be called as its advantages over other local
search methods. SA can be defined as versatile because it does not depend on any
restrictive features. SA algorithms are easily rearrangeable with code changes to
increase the performance of the optimization algorithm. However, SA algorithm needs
connections between solutions and time to calculate them. There may be sensitive
parameters for the calculations, and it significantly affects the quality of the solutions
(Göktürkler, 2011).

23
CHAPTER THREE
LITERATURE REVIEW OF GEOPHYSICAL METHODS BASED ON
METAHEURISTICS

A wide variety of the inversion problems are solved by metaheuristic methods in


recent years to reach the global minimum in geophysical studies. Because
metaheuristics are more suitable for best fitting model estimation compared to the
derivative-based techniques. Therefore, it is important to investigate the studies on
geophysical applications with metaheuristics for the performance analysis of an
algorithm, comparison studies among desired algorithms and proposing a new
algorithm.

To express the effects of metaheuristic methods on the performance of the inversion


studies, literature of the main metaheuristic methods is presented in detail for various
geophysical methods. For this purpose, geophysical applications of Genetic Algorithm
(GA), Particle Swarm Optimization (PSO), Simulated Annealing (SA), Differential
Evolution (DE) and Genetic-Price Algorithm (GPA) are investigated.

3.1 Genetic Algorithm Applications

Genetic Algorithm (GA) have attracted the attentions for inversions studies in
geoscience for many years with its direct and hybridized versions. GA offers a suitable
option for best fitting model parameter estimation of inverse problems in geophysics.
Therefore, GA is applied by the researchers on geophysical areas such as
electromagnetic, gravity, magnetic, magnetotelluric (MT), resistivity, seismic, self-
potential and vertical electrical sounding for the convergence of solution with higher
fitness values. The main applications of the GA for the various geophysical inverse
problems are listed as follows:

a. Electromagnetic Data: Pendock, (1993); Qing et al., (2001); Yogi and


Widodo, (2017); Ayani et al., (2017); Guo et al., (2019); Ayani et al., (2020)

24
b. Gravity Data: Montesinos et al., (2005); Qiu et. al., (2009); Amjadi and Naji,
(2013); Osman and Alabora, (2015); Mahbuby et al., (2017)
c. Magnetic Data: Currenti et al., (2007); Akca et al., (2014); Montesinos et al.,
(2016); Kaftan, (2017)
d. Magnetotelluric Data: Moorkamp et al., (2007); Tong et al., (2009); Wu et al.,
(2018); Li et al., (2018); Wang et al., (2018); Wushouaili et al., (2018)
e. Resistivity: Schwarzbach et al., (2005); Madan et al., (2008); Liu et al., (2012);
Attwa et al., (2014)
f. Seismic: Soupios et al., (2011); Li et al., (2013); Aleardi, (2015); Rubaiyn et
al., (2019); Xing and Mazzotti, (2019a); Xing and Mazzotti, (2019b); Qin et
al., (2020);
g. Self-Potential Data: Abdelazeem and Gobashy, (2006); Gokturkler and
Balkaya, (2012)
h. Vertical Electrical Sounding: Madan et al., (2008); Akca et al., (2014)

Genetic Algorithm was used for the estimation of a distribution of conductivity


from airborne electromagnetic data over the South Africa by Pendock, (1993). The
problem was first defined as a Bayesian inference problem and then the observed data
were estimated by GA. The results indicated that GA is efficient for estimating high
dimensional parameters with low computer storage requirement. Qing et al., (2001)
used real coded Genetic Algorithm for the reconstruction of conducting cylinders with
random noise. Before the inverse problem was solved by the real coded Genetic
Algorithm, the problem had considered as a restrained minimization problem. The
performance of the real coded Genetic Algorithm was found more accurate than
standard Genetic Algorithm. An inversion study of 1D time domain electromagnetic
synthetic data was performed by Yogi and Widodo, (2017) using both Genetic
Algorithm and Particle Swarm Optimization. The results of the parameter estimations
of the electromagnetic data were similar for trend, value, and computation time. A
frequency domain marine controlled source electromagnetic data was inverted by
Ayani et al., (2017) with Genetic Algorithm. This study used Bayesian framework for
the combination of the model parameters with forward problem for the estimation of
the probability of density function. The study was carried out for both contaminated

25
with pseudo random numbers and noise free data. Reliable estimations of the thickness
of inline electric field were achieved by the Genetic Algorithm. An inverse problem
of electromagnetic tomography was solved by Guo et al., (2019) using a developed
version of GA named adaptive multi population parallel Genetic Algorithm. The
developed algorithm is verified by a numerical experiment and field tests the results
show that good and realistic estimation parameters were achieved by improving global
and local search capabilities of the Genetic Algorithm. Ayani et al., (2020) studied an
inversion problem of marine controlled source electromagnetic data. A non-dominated
sorting Genetic Algorithm was developed for the inversion of these problem. A
contaminated with pseudo random numbers synthetic data sets to were used to
demonstrate the accuracy of the study. According to the results of the study, proposed
Genetic Algorithm was feasible for 1D problems for robust estimations and low
computational costs.

Genetic Algorithm implemented by Montesinos et al., (2005) for 3D inversion of


gravity data located in Canary Islands. The geometry of the sources of the anomaly
gravity field were determined by three operators of the Genetic Algorithm as selection,
crossover and mutation. A synthetic data and a field data form the Canary Island used
for the confirmation of the efficiency of the algorithm for the inversion of gravity data.
The results yielded accurate solutions and clear correlations between the gravity field
in the model and volcanic structures of Fuerteventura in Canary Island. An estimation
the model parameters of synthetic gravity data were carried out by Qiu et al., (2009)
using Genetic Algorithm and generalized least squares. Desired global optimization
solution was found by the inversion of gravity data with rapid initial convergence. On
the other hand, it is indicated as an advantage that setting of initial parameters were
not necessary in the Genetic Algorithm. A depth determination study from residual
gravity anomalies was aimed by Amjadi and Naji (2013) using a combined method
with Genetic Algorithm and Least Square Method. The proposed method offered to
prevent premature convergence to the local solutions and low accuracy of the solution.
The validation of the study was tested with synthetic data with random errors.
Additionally, the method was applied to a real gravity data from USA. Finally, a good
agreement was obtained by the combination of the Genetic Algorithm and Least

26
Square Method. A new modeling technique was proposed by Osman and Albora
(2015) about modeling of gravity anomalies due to 2D geological structures using
Genetic Algorithm. Two synthetic data were tested for the validation. Results of the
modeling with GA were compared with the results from the literature. Low values of
the mean errors were found in the tests and the GA was defined as reliable for this
technique. Local gravity modeling was studied by Mahbuby et al., (2017) using
spherical radial basis functions and a genetic algorithm. Low rms values were found
as differences between predicted and observed gravity anomalies from the numerical
tests using Genetic Algorithm.

Magnetic data of Etna volcano were studied by Currenti et al., (2007) for the
inversion using Genetic Algorithm. Field data sets of these magnetic anomaly was
recorded at 2002-2003 Etna flank eruption. The magneto-elastic parameters were
estimated using GA in the upper crust on Mt. Etna. The results show that the model
parameters of the deformation pattern and the magnetic anomalies were consistent
with the real data. Inversion of the magnetic resonance sounding, and vertical electrical
sounding were carried out by Akca et al., (2013) for the model parameter estimation
with multi objective Genetic Algorithm. A synthetic data was used for the tests of the
multi objective optimization problem. On the other hand, verification of the inversion
problem was performed by borehole data. The application of multi objective GA cause
effective and efficient solution for exploring the search space. Montesinos et al. (2016)
applied Genetic Algorithm on inverse problem of magnetic anomaly from Canary
Islands to estimate 3D geometry and location parameters and according to the
inversion results with Genetic Algorithm. The magnetization vectors including
intensity, inclination and declination, and magnetic field direction of the Earth were
considered as inputs. Two complex synthetic anomalies were inverted by GA and the
inversion results were found close to the real values. Kaftan, (2017) investigated
inversion of magnetic anomalies over thick dike using GA. Model parameters of
synthetic and field magnetic anomalies (Arizona, USA and Bayburt, Türkiye) such as
depth to the top of the dike, the half-width of the dike, the dip of the thick dike, and
the susceptibility contrast were determined by GA. Additionally, the synthetic
anomaly was tested as noise-free and contaminated with pseudo random numbers

27
magnetic data. At the end of the inversion study for magnetic anomalies, GA was
defined as a useful tool for evaluating magnetic anomalies.

The compatibility of long-period MT data and P-wave teleseismic receiver


functions were examined by Moorkamp et al., (2007) using Genetic Algorithm for
determination of compatibility between seismic velocities and electrical
conductivities. Teleseismic and MT data were taken from the Slave craton for the
inversion study. According to the inversion method, compatibility of electrical and
seismic structures has been achieved with Genetic Algorithm based approach. Tong et
al., (2009) suggested a hybrid GA for a MT inverse problem. The hybrid GA was
developed as the combination of simplex method and Genetic Algorithm. This
modification was aimed to increase local search ability, numbers of calculation and
adaptability for large space of standard Genetic Algorithm. For this purpose, a typical
multi-solution function of Schwefel as a synthetic MT data and a simple earth model
was used for testing the proposed optimization method. The test results showed good
convergence, avoiding earliness, and improving precision. Li et al., (2018) proposed
Genetic Algorithm for MT noise suppression. The adaptability of searching the
optimal matching atom and the location were completed by Genetic Algorithm.
Estimated and measured MT data were compared and optimal matching atoms with
different kinds of noise interference was ensured by the GA. Wang et al., (2018)
studied optimization of MT inversion problems based on back propagation neural
network using Genetic Algorithm. Operation efficiency and calculation accuracy were
increased in the tests on synthetic and real data with the Genetic Algorithm.
Wushouaili et al., (2018) discussed the linearization method of MT symmetric
anisotropic layer using Genetic Algorithm. Test results of Genetic Algorithm with
noisy theoretical model yielded feasible and simple solution compared by true results
of the geoelectric profile. Wu et al., (2018) investigated a joint inversion study for the
determination of the model parameters of 1-D MT and Surface-Wave Dispersion Data
using an improved multi objective GA. Two numerical tests were performed to
determine the advantages of the algorithm. The results indicated that observed
parameters from Longmenshan fault zone, Tibetan Plateau and Sichuan Basin
reflected the properties of the areas truly with Genetic Algorithm.

28
Schwarzbach et al., (2005) studied Genetic Algorithm for the 2D inversion of direct
current resistivity data. Plausible results have been recorded after the inversion of the
multi-objective minimization optimization with restricted population size and stopping
criterion because of the long computational times. An assessment of groundwater
condition using surface resistivity and vertical electrical sounding inversion was
carried out by Madan et al., (2008) with Genetic Algorithm. The purpose of the study
was declared to reveal a resistivity contour map of the groundwater for different depths
by estimating the model parameters with GA. As a result, optimal model parameters
were found as efficient and reliable with low values of root mean square error. An
improved Genetic Algorithm was implemented by Liu et al. (2012) for the inversion
of 3D resistivity. The Genetic Algorithm was improved with the mutation direction
control method for better search direction. Therefore, search efficiency was increased
with the optimization of the mutation direction. Thus, high quality inversion results
have been obtained by the inversion of synthetic and practical examples with the
improvement on the control method of mutation direction. An application of Genetic
Algorithm was performed by Attwa et al., (2014) for the 1D inversion of soil
characterization using DC resistivity data. Data of the inversion was used from the
near surface environmental status of soil in Tenth of Ramadan city, Egypt. At the end
of the optimization of the inversion study, the subsurface imaging of the resistivity
distribution was achieved by accurate and quick solution from the GA.

Soupios et al. (2011) applied hybrid Genetic Algorithm which was include least
squares and Genetic Algorithm, for the parameter estimation of a seismic tomography.
Synthetic refraction travel-time date sets and real data from a crosswell tomographic
project were used for the inversion of the seismic tomography. As a result, robust and
reliable tomographic images were obtained by solving large dimensional problems
with and achieving fast location to global minima with hybrid Genetic Algorithm.
Inversion of Rayleigh wave dispersion curve with Genetic Algorithm was investigated
by Li et al., (2013) for the design of a software programming. Accurate inversion
solution was obtained with Genetic Algorithm by avoiding local optimal solutions.
Aleardi (2015) compared the results of seismic velocity estimation from Genetic

29
Algorithm, neural networks, and multilinear approaches. A series of blind tests and
evaluations of the correlation coefficients were performed for the effectiveness of the
methodology based on the estimating the data in missing parts of velocity logs. Robust
results were achieved by Genetic Algorithm compared with other methods at nonlinear
relations between the input logs and the seismic velocity. Rubaiyn et al., (2019) used
Genetic Algorithm for Inversion of Rayleigh Wave Dispersion Curve to estimate the
model parameters of near surface S-wave. The results of the shear-wave velocity
profiles from the inversion of two synthetic models were compared with the true
model. Additionally, a real-world data from the Institut Teknologi Bandung campus
area were used to find inverted dispersion curve and vertical shear-wave velocity
profile using GA. As a results of the tests, GA approach was found as effective to
estimate the shear-wave velocity profile from Rayleigh wave. Xing and Mazzotti,
(2019a and 2019b) studied Rayleigh wave inversion using GA in two separate works
as method and synthetic examples and application to two actual data sets. First part of
the study included a development of new method called two-grid Genetic Algorithm
for the inversion problem. Three synthetic data sets were used for the inversion. Good
agreements between observed and predicted values were achieved with suggested
algorithm in all tests. On the other hand, two actual data sets two actual data sets from
Luni (Italy) and Grenoble (France) were used to Rayleigh wave inversion with two-
grid Genetic Algorithm. The results showed that, fair inversion outcomes were yielded
with few mismatches between the predicted and observed traces. Qin et al., (2020)
investigated inversion of two geophysical techniques named interactive integrated
interpretation of Ground-penetrating radar and Rayleigh wave data with Genetic
Algorithm to reconstruct the near-surface distribution. According to the results of the
field experiments, Genetic Algorithm were described as efficient and effective method.

Abdelazeem and Gobashy (2006) presented genetic algorithm as a new technique


for the inversion of self-potential anomalies for simple geometrical structures.
Inversion study was aimed to find model parameters of unknown depth, polarization
angle, and shape factor with GA. Noise-free and noise-contaminated synthetic data
and field data sets from Germany and Türkiye were tested with GA to find inverted
parameters. As a result of these study, realistic results of the root-mean squared error

30
have been reached by genetic algorithm for both synthetic and field data sets of the
self-potential data. Göktürkler and Balkaya (2012) used Genetic Algorithm for the
inversion of self-potential anomalies to estimate model parameters of both synthetic
and field data sets and compare the results with Simulated Annealing and Particle
Swarm Optimization. Synthetic data of self-potential anomalies with simple
geometries were inverted as both without noise and noise-contaminated. On the other
hand, three different field anomalies of copper belt (India), graphite deposits
(Germany) and metallic sulfide (Türkiye) were inverted to estimate the model
parameters using GA. The model parameters were defined as electric dipole moment,
polarization angle, depth, shape factor and origin of the anomaly. Similar consistent
results from Genetic Algorithm were found for the global minimum with other
algorithms.

Genetic Algorithm was studied by Madan et al., (2008) for the inversion of both
vertical electrical sounding and resistivity, which was mentioned before in resistivity
literature. Subsurface layer parameters as true resistivity and thickness from the
vertical electrical sounding data were optimized with GA based inversion study.
Genetic Algorithm expressed as efficient and reliable for determining model
parameters of the vertical electrical sounding data.

3.2 Particle Swarm Optimization Applications

Particle Swarm Optimization (PSO) have been utilized especially since last decade
in wide range of geophysical fields for model parameter estimations by inversion. PSO
offers a successful alternative for the inversion in geophysics with the benefits of
stability, accurate tuning possibility, low computational costs, and suitability for
multiple data sets. Therefore, PSO is applied in many geophysical fields such as
electromagnetic, gravity, magnetic, MT, resistivity, seismic, self-potential and vertical
electrical sounding for consistent solutions. The main applications of the PSO for the
various geophysical inverse problems are listed as follows:

a. Electromagnetic Data: Godio and Santilano (2018); Li et al., (2019)

31
b. Gravity Data: Pallero et al., (2017); Jamasb et al., (2019); Essa and Munschy
(2019); Essa and Geraud (2020); Anderson et al., (2020); Essa et al. (2021);
Pallero et al., (2021); Roy et al., (2021)
c. Magnetic Data: Srivastava and Agarwal (2010); Jie and Tao (2015); Essa and
Elhussein (2018); Essa and Elhussein (2020), Heidari et al., (2022)
d. Magnetotelluric Data: Shi et al., (2009); Patel et al., (2016); Grandis and
Maulana (2017); Darisma and Marwan (2019); Karcıoğlu and Gürer (2019);
Pace et al., (2019); Cui et al., (2020)
e. Resistivity: Pekşen et al., (2014); Cui et al., (2017); Yan et al., (2020)
f. Seismic: Yue et al., (2009); Tronicke et al., (2011); Poormirzaee et al., (2014);
Poormirzaee et al., (2015); Poormirzaee (2016)
g. Self-Potential Data: Sweilam et al., (2007); Santos (2010); Pekşen et al.,
(2011); Göktürkler and Balkaya (2012); Singh et al., (2015); Srivardhan et al.,
(2016); Essa (2019); Essa (2020)
h. Vertical Electrical Sounding: Martinez et al., (2010); Ramadhany et al., (2019)

An inversion method was designed by Godio and Santilano (2018) using PSO
algorithm to estimate model parameters of electromagnetic geophysical data.
Performance of the PSO was determined by sensitivity analysis. According to the tests
with synthetic data, PSO algorithm was proved its success for the inversion of complex
and multi parameter problems. Li et al., (2019) proposed a PSO based optimization
algorithm, which was designed at combination of PSO and damped least squares
algorithm, for the inversion of 1D transient electromagnetic data. The inversions of
synthetic and field data sets were performed using proposed algorithm. The results
show that combined PSO based algorithm yielded high optimization efficiency and
consistent with the logging data.

PSO was investigated by Pallero et al., (2017) for the inversion of 3D nonlinear
gravity anomaly and uncertainty assessment. Values of true basin depths were
compared with the observed points of without noise and noise-contaminated synthetic
data of sedimentary basins. Parameter tuning study was carried out for PSO according
to the synthetic inversion studies. Then, inversion of real gravity data from Argeles

32
Gazost in French Pyrenees was performed using PSO algorithm. Finally, PSO was
defined as a very good alternative for the model parameter estimation of 3D gravity
anomalies and uncertainty assessments. Jamasb et al., (2019) was proposed a hybrid
PSO based optimization algorithm including evolution strategies to reduce
computation time. The proposed algorithm was used to inversion of a real 3D nonlinear
gravity anomaly from South Caspian Basin for the estimation of the thickness of the
sedimentary cover. The results of the inversion with the PSO based hybrid algorithm
yielded meaningful solution and good agreement with measurement values without
premature convergence. Essa and Munschy (2019) developed a new method for the
inversion of second moving average residual gravity anomalies using PSO algorithm.
Model parameters of buried structure as amplitude factor, depth, location, and shape
were estimated by PSO from the inversion of two theoretical and three real data sets
from Cuba, Canada and India. The results of model parameters of field examples from
presented study were compared with the results from previous literature. Good
agreements were obtained with previous studies using PSO and efficiency of the
algorithm was confirmed by the inversion of theoretical and field data sets. Essa and
Geraud (2020) investigated PSO for the model parameter estimations of two-
dimensional horizontal thin sheet parameters by inversion of gravity anomaly with
second moving average method. Estimation parameters of depth, width, thickness,
density contrast and origin location of the anomaly were found by the inversions of
without noise, noise-contaminated synthetic and real gravity data from three different
locations about oil, gas, and mineral exploration. The results from the inversions using
PSO were obtained as applicable for especially real data sets with very short
computational time. Anderson et al., (2020) proposed PSO algorithm for inversion of
gravity anomaly from 2D vertical fault structure. PSO was used to invert three types
of synthetic data as residual, regional and composite gravity anomaly and two different
field data sets from Egypt as Gazelle fault and Mersa Matruh fault. As a result of the
inversion studies with PSO, robustness of the inversion was seen for both synthetic
and field gravity data. Essa et al., (2021) inverted a two-sided fault gravity anomaly
using PSO algorithm to determine model parameters of the upthrown and downthrown
of the fault. Various without noise and noise-contaminated synthetic examples were
investigated for the verification and assessment of the accuracy. On the other hand,

33
two field data were successfully inverted with PSO from India and USA. Estimated
model parameters of the inversion results were compared with the results reported in
the previous literature and good agreement were observed in the parameter results
using PSO. Roy et al., (2021) used PSO for the inversion study to estimate the
parameters of depth of sedimentary basin of residual gravity anomalies. Noise-free and
noise-contaminated synthetic models were used in the inversion and uncertainty
appraisal analysis. Additionally, PSO was used for the gravity inversions of two field
data as Godavari basin from India and Sayula basin from Mexico. As a result, a small
number of model parameters were successfully estimated by PSO with low
computational burden. Pallero et al., (2021) developed a MATLAB tool using PSO
algorithm for the inversion of 2D gravity anomaly due to sedimentary basins. The
MATLAB tool based on the PSO for the gravity inversion studies included model
definitions, inversions and result processing for model parameter estimation,
uncertainty analysis and plot representations. PSO based MATLAB tool was offered
gravity regional trend estimation, vertically and horizontally density contrast
variations for both synthetic and field gravity anomalies.

Srivastava and Agarwal (2010) proposed PSO algorithm for inversion of magnetic
anomaly profile due to amplitude of 2D analytic signal. The amplitude values of
magnetic anomalies were interpreted by the inversion studies of without noise and
noise-contaminated synthetic data with PSO. Parameter estimation results showed
high accuracy and stability compared to other methods by the inversion of PSO. Jie
and Tao (2015) inverted two-dimensional magnetic data using multi objective PSO
inversion with the purpose of simultaneously minimization of data misfit and model
constraints. Therefore, multi objective inversion solution was obtained without
gradient information of the objective function. Feasible solutions were found by the
inversion of synthetic 2D magnetic data by PSO based proposed algorithm. Essa and
Elhussenin (2018) carried out an inversion study by PSO for magnetic anomalies of
simple geometrical bodies. The inversion study was aimed to estimate model
parameters of depth of body, amplitude coefficient, angle of effective magnetization,
shape factor and horizontal coordinates of the body. Root means square error was
determined for the evaluation of misfit between observed and calculated anomalies.

34
Synthetic data without noise, noise-contaminated synthetic with random noise levels
and two field data sets from USA and Egypt were inverted by PSO algorithm. Finally,
the inversion results showed good agreement with the results from the previously
published literature. Another study of Essa and Elhussenin (2020) was focused
inversion of magnetic data from mineral exploration using PSO algorithm. After the
utilization of PSO for the inversion of without noise and noise-contaminated synthetic
data, three different field examples from Canada, Türkiye and Egypt were investigated.
As an inversion result, estimated parameters were found close to the known values
with PSO. A current is performed by Heidari et al., (2022) for the parameter estimation
of depth and shape of magnetic anomaly using the improved global PSO. The study
area is located Ileh region in Iran and the area is rich for iron resources. As a result of
the study, improved global PSO is defined as capable for the model parameter
estimation with high accuracy.

MT data inversion was performed by Shi et al., (2009) using damped PSO which
was proposed for the optimization of multi-dimensional problems and dynamic target
searching. Proposed PSO based algorithm was tested with synthetic data. Inversion
results indicated that proposed PSO algorithm was fast and effective for MT
anomalies. Patel et al., (2016) developed a hybrid algorithm with PSO and
gravitational search algorithm for 1D inversion of MT data. Proposed PSO based
hybrid algorithm was tested by without noise and noise-contaminated synthetic data
and one field data from India. Available results as global optimum were achieved with
fast convergence using proposed algorithm. Grandis and Maulana (2017) applied PSO
to 1D inversion modeling as simple minimization problem for synthetic MT anomalies
with various layers. According to the inversion with PSO, satisfactory results were
obtained for synthetic MT data. Darisma and Marwan (2019) carried out 1D MT
inversion study using PSO algorithm. Three-layer model of without noise and noise-
contaminated synthetic with 10% Gaussian noise MT data were inverted by PSO to
estimate global best solutions of model parameters. Additionally, inversion results
from PSO were compared with the results of Levenberg-Marquardt algorithm.
Accurate model parameter estimation results were achieved and both algorithms were
defined as successful for determination of actual model. Karcıoğlu and Gürer (2019)

35
implemented PSO algorithm to an inversion of 2D smooth modeling of radio MT data.
PSO was used to minimization of change vector which was required to smoothness of
the of the model parameters because of avoiding multi objective optimization. This
approach was tested by the inversion of synthetic and a field data set from Kerpen area
in Cologne, Germany with PSO algorithm. Successful global minimization of the
change vector was achieved, and this approach was suggested for the smooth
inversions. Pace et al., (2019) was also applied PSO algorithm to the 2D MT inversion
studies. First, the validation of the PSO algorithm was investigated with two synthetic
models. Then, PSO was applied for the inversion of a real data from Canada.
Hierarchical PSO with time varying acceleration coefficients were applied for the
better stability and convergence of the inversions. Additionally, PSO code was
parallelized for short computational time. After the modifications of the PSO
algorithm, 2D MT data was successfully inverted. Cui et al., (2020) studied inversion
of MT data with a PSO based hybrid algorithm, which was the combination of PSO
and regularized least squares method. According to the operation logic of the hybrid
algorithm, PSO was used to preliminary search the model parameter space to
determine the initial model for iteration. Then, the inversion was continued with the
least square regularization. The inversions of both synthetic and field data sets of MT
data yielded good performance with high efficiency and accuracy with PSO based
hybrid optimization algorithm.

Yue et al., (2009) proposed PSO algorithm for wavelet inversion of pre-stack
seismic extraction. Seismic wavelet vector was implemented the algorithm in a form
of particle for better inversion results. On the other hand, three-point mean filter was
used to update particle velocity for quick solution of the optimization. Real pre-stack
seismic data was successfully inverted with the improved PSO algorithm. According
to the inversion results, PSO was suggested for efficient and feasible inversion of
seismic wavelet. Tronicke et al., (2011) used PSO algorithm for joint inversion of GPR
and P-wave seismic travel times. The inversion of a synthetic data was performed
using PSO and the results were presented for zonal model, GPR velocity and P-wave
velocity. The comparison of inversion results with the true values showed that PSO
algorithm was capable for the seismic inversions with high convergence rates.

36
Poormirzaee et al., (2014) investigated inversion of refraction seismic data using PSO
algorithm. Various synthetic models were inverted by PSO to determine the efficiency
of the algorithm. Additionally, inversion of a field data from Tabriz in Iran was
performed. Finally, the inversion results with low misfit and acceptable convergence
speed were obtained for refraction seismic data with PSO algorithm. Poormirzaee et
al., (2015) performed a similar study about the inversion of seismic refraction using
PSO algorithm as a case study for the investigation of field anomaly from Tabriz in
Iran. The inversion results with plausible misfit and convergence speed showed that
the study area was composed of two layers and properties of the layers were calculated
for layer velocities and thicknesses of two layers. Poormirzaee (2016) was also applied
PSO algorithm for S-wave velocity profiling by the inversion of refraction
microtremor data. Two noise-contaminated synthetic, two without noise synthetic and
a field data from Tabriz in Iran were investigated for the inversion with PSO.
Estimated S-wave velocity profile was compared with downhole data of the area and
the results confirmed PSO was a suitable algorithm for the inversion of refraction
microtremor data.

Pekşen et al., (2014) examined an inversion performance of one-dimensional DC


resistivity due to two-layered anisotropic earth model using PSO algorithm. Model
parameters of horizontal and vertical resistivities, and thickness from anisotropic earth
model were estimated by PSO. Synthetic without noise, noise-contaminated synthetic
with 5% Gaussian noise and a field data from Türkiye were inverted by PSO with high
accuracy and 95% confidence interval of probability density function. Cui et al.,
(2017) studied PSO algorithm for the joint inversion of resistivity and induced
polarization-sounding data. Data of joint inversion were decomposed as resistivity and
equivalent resistivity data. Both decomposed data were inverted simultaneous and
sequential. According to the inversion results of numerical examples and comparison
results of the results from generalized linear method, PSO was yielded solutions with
high accuracy, fast convergence, and good posterior probability distributions. Yan et
al., (2020) investigated ultra-deep azimuthal resistivity for inversion and uncertainty
assessment using PSO algorithm. Performance of the PSO for inversion studies were
determined by parameter analysis and parallelization. On the other hand, convergence

37
rate was tried to increase with a gradient method. Based on the inversions of several
synthetic data and a field data from Australia, PSO found successful about global
searching capability and simplicity.

PSO algorithm was applied to the self-potential signal inversion by Sweilan et al.,
(2007) for the estimation of model parameters of simple polarized bodies. Minimum
of an objective function was determined by PSO by the estimation of the parameters
as depth to the electric source, electric charge polarization angle, electric dipole
moment and shape factor of a buried body. Validation of the inversion with PSO was
tested with without noise and noise-contaminated synthetic data. Estimation results
showed good agreement with the true value and maximum error was observed 0.25 %
from PSO. On the other hand, three field data sets were inverted with PSO, and the
inversion results were found very close to the results of other metaheuristic methods.
Santos (2010) was investigated PSO algorithm for the inversion of self-potential
anomalies of simple geometry such as spheres, cylinders, and inclined sheets.
Estimated parameters from the inversion with PSO were listed as depth, exact origin,
dipole moment and polarization angle. PSO algorithm was tested with synthetic data
of single and multiple self-potential anomalies and two field data sets from Surda area
in India and Vilarelho da Raia in Portugal. Inversion results of the field anomalies were
compared other studies and PSO showed good agreement with fast convergence
ability. Pekşen et al., (2011) studied inversion of self-potential anomalies of simple
bodies for model parameter estimations. For the inversions, without noise and noise-
contaminated synthetic and three field data from Türkiye were performed using PSO
algorithm. As a result of the inversion, contour maps of low misfit areas, values versus
the number of generation and frequency distributions of each parameter were
presented for synthetic and field data sets and PSO was suggested for the inversion of
SP anomalies. Göktürkler and Balkaya (2012) was investigated the inversions of SP
anomalies caused simple geometries with various metaheuristic methods which was
previously mentioned. The results of the inversions of field anomalies with PSO
algorithm indicated that lowest rms error was observed from Bavarian data in Germany
and Süleymanköy data in Türkiye compared to other metaheuristic and local
optimization methods. Therefore, PSO yielded successful results and better CPU time

38
for the inversions of the SP anomalies. Singh et al., (2015) performed an inversion
study using PSO for SP anomaly to detect coal fires. For this purpose, first, the PSO
was tested with noise-contaminated synthetic SP anomaly, then the field anomaly from
Jharia Coal Field in India was inverted with PSO and the results were compared with
measured data. Finally, robust performance was achieved for PSO algorithm for the
inversion of SP anomaly. Srivardhan et al., (2016), carried out similar study for the
inversion SP anomaly of coal fires from Jharia Coal Field in India using PSO
algorithm. Depth estimation of coal fires was aimed by the inversions with PSO. The
estimation of desired parameter was achieved with good accuracy by the inversion
using PSO algorithm. Essa (2019) used PSO algorithm for the SP inversion to interpret
observed anomalies that measured along a profile. Inversion study was performed to
estimate the model parameters of amplitude coefficient, depth, polarization angle and
the shape factor for three synthetic and two field data from USA and Türkiye. The
PSO inversion results showed good agreement with the known parameters of
anomalies for synthetic data and other studies for field data. Essa (2020) investigated
SP inversion for finite 2D inclined dike using PSO algorithm. Synthetic and field data
sets were inverted with PSO to estimate the model parameter of polarization
parameter, depth, inclination angle, width, and location parameter for SP anomaly. As
a result of the study, inversion solutions of field data sets from Germany and India
were found and compared with the accessible values of field areas.

Martinez et al., (2010) proposed improved PSO algorithm to increase convergence


for the geophysical inverse problems and the algorithm was tested using synthetic
functions. Then, improved PSO algorithm was used for the inversion of vertical
electrical sounding field data from Spain. Inversion results of convergence curves and
posterior histograms of sea water intrusion depth were compared with the results of
binary genetic algorithms and simulated annealing. According to the comparisons,
higher convergence rate with lower misfit was achieved with the PSO algorithm.
Ramadhany et al., (2019) proposed regressive-regressive PSO algorithm for the
inversion of vertical electrical sounding data to determine the distribution of aquifer
zone from Pacitan Region in Indonesia. Finally, groundwater identification including

39
the determination of sand, gravel, clay, and aquifer zone in the desired area was
successfully achieved with the inversion using proposed PSO based algorithm.

3.3 Simulated Annealing Applications

Simulated Annealing (SA) have been used as metaheuristic method to find global
optimum point from the search space with integer variables in geophysics. SA is a
useful and suitable technique in geophysics to find to find a local minimum using an
approximate global optimum point. Thus, SA and its variation of Very Fast Simulating
Annealing (VFSA) are commonly preferred in many areas of the geophysical design
an optimization study for the characteristics of simple implementation and high speed
to reach the solution. The main applications of the SA for the various geophysical
inverse problems are listed as follows:

a. Electromagnetic Data: Sharma and Kaikonen (1999); Yin and Hodges, (2007);
Wang et al., (2018)
b. Gravity Data: Mundim et al., (1998); Nagihara and Hall (2001); Roy et al.,
(2005); Yu et al., (2007); Biswas (2015); Biswas (2016a)
c. Magnetic Data: Biswas (2016); Biswas and Acharya (2016)
d. Magnetotelluric Data: Chen et al., (2012)
e. Resistivity: Dittmer and Szymanski (1995); Chunduru et al., (1996); Sharma
(2012);
f. Seismic: Mosegaard and Vestergaard (1991); Sen and Stoffa (1991); Weber
(2000); Beaty et al., (2002); Ma (2002); Yu et al., (2007)
g. Self-Potential Data: Göktürkler and Balkaya (2012); Biswas and Sharma
(2014); Biswas and Sharma (2015); Biswas (2016b); Biswas and Sharma
(2017); Sharma and Biswas (2013);

Time domain electromagnetic data was studied by Sharma and Kaikkonen (1999)
to observe transient electromagnetic responses using VFSA as a global optimization
method. The inversion was performed with VFSA using both without noise and noise-
contaminated synthetic data. Additionally, a field data from South Australia was used

40
to compare the results of nine parameters of conducting subsurface between observed
and computed response. Reliable estimates of all model parameters were achieved by
the VFSA for the electromagnetic data. Another electromagnetic inversion study was
performed by Yin and Hodges (2007) for airborne with SA. Theoretical and field data
sets were inverted by SA for the examinations of the effectiveness of the algorithm
and comparison of the results with other algorithms. More robust and smoother
solutions were obtained by SA than the traditional algorithms. Wang et al., (2008)
investigated a transient electromagnetic inversion using simplex SA algorithm. Local
minimum value was obtained by the simplex method and then the SA algorithm was
used for the reach the global optimum solution. The inversion results of three different
stratigraphic models were compared for both noise-contaminated and without noise
conditions. According to the inversion results, simplex SA algorithm was useful option
for global searching ability and convergence speed.

Inversion of the gravity data for 2D and 3D density distributions was optimized by
Mundim et al., (1998) using generalized SA algorithm. SA algorithm was tested for
synthetic data of simple geometrical shapes in gravity inversion. The results were
indicated that SA enabled to reach to the optimal model parameters than conventional
approaches such as Boltzmann and Cauchy methods. Nagihara and Hall (2001) used
SA algorithm for 3D gravity inversion to locate and constrain the geometry of gravity
anomalies over the salt structures. A synthetic gravity data for the salt structures in the
Gulf of Mexico was used for the testing of the inversion method. It was expressed that
SA method was computationally efficient to solve the gravity inverse problems. Roy
et al., (2005) investigated both inversion and uncertainty estimation of gravity data
over Lake Vostok, East Antarctica using SA algorithm. It was aimed to find the model
parameters of the coordinates or shapes of 2D and 3D bodies with SA. On the other
hand, posterior probability density was found according to the samples that obtained
by SA. For the calculations, two synthetic data sets and one airborne-gravity data set
applied on the inversion method. Estimated model parameters and posterior
probability density values showed SA is a useful and successful alternative as global
optimization method. Yu et al., (2007) employed VFSA to the constrained joint
inversion of both gravity and seismic data. Theoretical model tests and field data

41
applications were performed to the estimation of the model parameters with VFSA.
The optimization algorithm was proved as effective and practicable especially for the
field data. Additionally, VFSA was found superior for the single inversion of the
gravity data. Biswas (2015) studied the inversion of the residual gravity anomaly of
simple shaped geometries using VFSA. As a results of the inversion, convergence
patterns and histogram plots of probability density functions were given for sphere,
horizontal cylinder, and vertical cylinder. VFSA was yielded reliable results with the
low uncertainty in short computation time for without noise and noise-contaminated
synthetic data and field examples. Biswas (2016a) also applied VFSA to the inversion
of gravity and magnetic anomaly over thin shaped structure. Without noise and noise-
contaminated synthetic data for single structures and a field data from Canada were
inverted with the VFSA algorithm. VFSA algorithm was found successful for the
estimation of the model parameters in short computational time and the agreement
with the calculated and actual parameters.

In the study of Biswas (2016a), mean value of the without noise and noise-
contaminated thin sheet type model parameters, convergence pattern of the model
parameters and misfit for magnetic data were presented as result of the inversion by
VFSA algorithm. Additionally, histogram plots and scatter plots of the model
parameters with probability density function were also given for the magnetic data.
VFSA yielded very good results without any uncertainty for magnetic data as well as
gravity. Biswas and Acharya (2016) implemented VFSA to the inversion of magnetic
anomaly caused by semi-infinitive vertical rod type structure. Three without noise and
noise-contaminated synthetic anomalies were inverted by VFSA to find the model
parameters of magnetic moment, location, depth and shape factor. Additionally, VFSA
was applied to a field data from Parnaiba basin, Brazil. Therefore, reliable results were
obtained for model parameters and uncertainty from the theoretical and field magnetic
data with VFSA.

Magnetotelluric-gravity joint inversion optimization was investigated by Chen et


al., (2012) with SA algorithm. According to the inversion results of theoretical and
practical data, good results were achieved by SA for the model disturbance,

42
inconsistency of magnetotelluric-gravity interfaces and weighted combinations of
objective functions.

Mosegaard and Vestergaard (1991) investigated SA algorithm to solve a seismic


trace inversion problem by determination of two-way travel times and reflection
coefficients. A field data from Denmark was used for the tests of SA algorithm. Closet
solutions to the true model were found by SA algorithm which was defined as capable
of producing reliable results. Sen and Stoffa (1991) employed SA for the inversion of
nonlinear one-dimensional seismic waveform. Without noise and noise-contaminated
synthetic data and a field data set from Carolina trough in USA were tested using SA.
Solutions were achieved by the inversion study close to the global minimum very
rapidly and SA algorithm defined as successful for seismic inversion. Weber (2000)
carried out seismic travel time tomography inversion using SA approach. Synthetic
without noise and noise-contaminated data sets and field data from Hungarian coal
mine were used for the inversion with SA. Reliable results with high convergence of
the solutions were achieved by SA algorithm according to the inversion results. Beaty
et al., (2002) was used SA for the inversion of Rayleigh wave dispersion curves with
fundamental and higher-mode for an S-wave velocity profile. According to the
inversions of synthetic and experimental data, SA was described as useful for
estimating velocity profiles from dispersion curve data. Ma (2002) was developed a
new version of SA to estimate acoustic and shear impedances from P-wave reflection
seismic data. Inversion studies carried out for global minimum of the objective
function and misfit between synthetic and observed seismic data. Inversion of the
synthetic and field data sets were successfully performed using SA algorithm which
was defined a suitable tool for the seismic data. Yu et al., (2007), which is mentioned
above in this section, was investigated seismic data in the joint inversion of gravity
and seismic using VFSA algorithm. VFSA was successful in the joint inversion with
high computing efficiency.

Resistivity data was studied by Dittmer and Szymanski (1995) for the inversion of
for both linear magnetics and non-linear resistivity problems. Synthetic vertical
profiling data is used to test the resistivity using SA. An archaeological field data from

43
Fountains Abbey in England was preferred for the model parameter estimation.
Therefore, SA algorithm was defined as a powerful method for the resistivity inversion
with high computing power. Chunduru et al., (1996) was applied VFSA to the
inversion of 2D dipole-dipole resistivity data for imaging the subsurface. Mean square
error was used for the minimization between synthetics and original data. Then, real
data from sulfide zone at Safford, Arizona was inverted, and the results was compared
with a model based on geologic and well-log data. A good agreement was seen
between the VFSA results and published results. Sharma (2012) introduced a program
using VFSA for 1D DC resistivity sounding data from various electrode arrays.
Synthetic and field data of resistivity sounding over different subsurface structures
were optimized presented VFSA program. The results were achieved with small
uncertainty and high probability density functions by VFSA algorithm.

Göktürkler and Balkaya (2012) inverted synthetic and field data tests from copper
belt (India), graphite deposits (Germany) and metallic sulfide (Türkiye) of SP anomaly
using SA to compare with PSO and GA, which was mentioned above in previous
subsections. Similar rms results of the model parameters were obtained using SA
compared to PSO and GA for both synthetic and field data sets. According to the
results, comparisons with other metaheuristic method and previous studies showed that
SA were able to provide efficient and compatible solutions with other optimization
algorithms. Sharma and Biswas (2013) implemented VFSA to the for the inversion of
self-potential anomaly of 2D inclined structure. Estimated model parameters of the
self-potential anomaly was aimed to find electric current dipole density, horizontal and
vertical locations of the center and half-width of the structure, and
polarization/inclination angle of the 2D inclined sheet by VFSA. Various synthetic
data and three field data from Surda in the area of Rakha Mines located in India and
the Bavarian woods, Germany was used for the inversion using VFSA algorithm.
VFSA was found more advantageous than other algorithms according to the solutions
and computational time. Biswas and Sharma (2014) studied inversion of SP anomaly
of 2D inclined sheet type structures using VFSA algorithm and analysis of ambiguity.
Ambiguities of the SP anomaly was observed for various approaches to find smallest
value of uncertainty. The inversion results indicated that VFSA was found useful for

44
SP anomalies with large number of variables. Biswas and Sharma (2015) used VFSA
for the inversion of SP anomaly from idealized bodies as sphere, horizontal and
vertical cylinder, and ambiguity analysis. This approach was tested by without noise
and noise-contaminated synthetic data and three field data sets. As a result of the study,
reliable estimation values of the model parameters were achieved by VFSA algorithm.
Biswas (2016) compared the performance of least square method and VFSA according
to the inversion of SP anomaly of 2D inclined sheet type structure. The comparative
study was performed for synthetic data without noise, noise-contaminated and two
field data. The results indicated that VFSA yielded results more precisely than linear
inversion technique with better computational time. Biswas and Sharma (2017) carried
out an inversion and uncertainty study with VFSA for SP anomaly over 2D inclined
thick sheet structures. According to the inversions of synthetic and field data, VFSA
was defined as useful for thick sheet type structure and for multiple bodies with short
computational time.

3.4 Differential Evolution Applications

Differential Evolution (DE) have been implemented to geophysical to reach global


optimum for real valued optimization problems without good initial parameters. DE is
a robust, simple, and easy to use metaheuristic method for solving optimization
problems. The main applications of the DE for the various geophysical inverse
problems are listed as follows:

a. Electromagnetic Data: Wang et al., (2010); Li et al., (2020); Li et al., (2021)


b. Gravity Data: Ekinci et al., (2016); Ekinci et al., (2019); Zeng et al., (2020);
Song et al., (2021);
c. Magnetic Data: Balkaya et al., (2017); Ekinci et al., (2017); Ekinci et al.,
(2019)
d. Magnetotelluric Data: Dong et al., (2015); Xiong et al., (2018); Zeng et al.,
(2020)
e. Resistivity: Li et al., (2020)

45
f. Seismic: Song et al., (2013); Song et al., (2014b); Zhu et al., (2015); Pan et
al., (2018); Riberio et al., (2019)
g. Self-Potential Data: Li and Yin (2012); Balkaya (2013)
h. Vertical Electrical Sounding: Balkaya (2013)

Wang et al., (2010) studied DE and its application to inversion of electromagnetic


imaging of heterogeneous media using dual frequency electrical conductivity imaging
equations. Two new regularizations algorithm was proposed to improve the rate of DE
convergence and reduce computation costs as Adaptive Recursive DE and Particle
Swarm Optimization DE. Finally, DE was defined as suitable to solve the problem of
electromagnetic inversion studies in geophysics. Transient electromagnetic inversion
was carried out by Li et al., (2020) using neural network and DE algorithm. An
improved approach for better global optimization ability was proposed using a chaotic
mutation and crossover with constraint factor DE. The results of the validation of
proposed DE based algorithm with two test functions and two geoelectric models’
inversion showed that DE based algorithm was feasible in electromagnetic inversion
applications according to the inversion speed, accuracy and stability. Another transient
electromagnetic inversion study was investigated by Li et al., (2021) using PSO and
DE algorithm. A joint algorithm was proposed based on PSO and DE for better global
optimization. As a result of the DE based proposed algorithm, inversion accuracy,
efficiency and stability were improved, and fast convergence was achieved with high
fitting degree in transient electromagnetic inversion.

DE was used first time in geophysics by Ekinci et al., (2016) and it applied on
residual gravity anomalies of simple shaped bodies for model parameter estimations.
Results of without noise, noise- contaminated synthetic single gravity anomalies, some
multiple source anomalies and three field anomalies were considered for the model
parameter estimations with DE algorithm. Uncertainty of the results were evaluated
by Metropolis-Hastings algorithm for both synthetic and field data sets. The inversion
results indicated that DE algorithm offered good accuracy and less computational
costs. Ekinci et al., (2019) also performed parameter estimation studies using DE to
compare with the PSO for inversion of gravity and magnetic anomalies. Inversions

46
were achieved for both synthetic and field data sets of gravity and magnetic anomalies
caused by deep seated fault structures. Synthetic anomalies were used for the
parameter tuning studies to determine optimum control parameters of the algorithm.
Field data sets from Graber oil field in USA and Perth Basin in Australia were inverted
by DE for gravity and magnetic anomalies respectively. Finally, DE was yielded great
results for robustness, consistency, computational cost, and convergence rate of the
algorithm. Zeng et al., (2020) proposed an improved DE algorithm for joint inversion
of MT and gravity data. Two population schemes and wide range petrophysical
constraints technology were based for the improvement of the DE. Fast convergence
speed and high optimization capability was achieved by DE after the joint inversion of
MT and gravity data. Inversion of gravity data was studied by Song et al., (2021) using
adaptive DE algorithm. The inversion was aimed to find physical property distribution
of gravity anomaly, which was occurred by causative sources embedded in the
subsurface. The adaptive characteristic of DE was acquired by the control of Lp-norm
regularization term. The results of the inversions of synthetic and field anomalies
showed that DE was useful for the gravity anomalies.

Balkaya et al., (2017) investigated DE algorithm for 3D nonlinear inversion of


magnetic anomalies caused by vertical sided prismatic bodies. Efficiency of the DE
algorithm was determined by the estimation of model parameters of synthetic and field
magnetic anomalies. Airborne magnetic anomalies from Biga Peninsula, Türkiye was
used as field data. DE was proposed as useful tool for the inversion of magnetic
anomalies based on the accuracy of the results. Ekinci et al., (2017) used DE algorithm
for 2D analytic signal amplitude inversion of total field magnetic anomalies. Small
scale magnetic anomaly from an iron deposit and large-scale magnetic anomaly from
a deep-seated magnetized structure were inverted for determination of the model
parameters as depths, geometries, and exact origins of the source bodies. DE algorithm
yielded reasonable solutions for 2D magnetic anomalies and low-dimensional data
inversions.

MT sounding data was inverted by Dong et al., (2015) using DE algorithm for the
extraction of IP information. The researchers preferred DE algorithm because of the

47
small effects of the initial model of DE to the inversion results. The results of the
inversion study were compared with the PSO. As a result of the comparison, DE
algorithm yielded better convergence rate than PSO. Xiong et al., (2018) proposed an
improved DE algorithm for the inversion of 1D layered geoelectrical synthetic MT
data. Consistent results were achieved by the inversion of without noise and noise-
contaminated synthetic MT data. On the other hand, numerical experiments were
performed for the investigation of DE over MT inversion. Finally, global exploration
and anti-noise capability were found effective for the numerical MT inversion. As
mentioned before, Zeng et al., (2020) studied the joint inversion of MT and gravity
data using an improved DE algorithm. According to the inversion results of fast
convergence speed and high optimization capability, DE algorithm was described
successful for wide range of physical constraint in joint inversions.

Song et al., (2013) applied DE algorithm to inversion of without noise and noise-
contaminated micro seismic data. Test results showed that good solutions were
obtained by DE according to the approximation degree to reach the exact solution.
Song et al., (2014b) investigated DE algorithm for nonlinear inversion of high-
frequency Rayleigh wave dispersion curves for near-surface S-wave velocity profiles.
After the evaluation of efficiency and stability of DE with the inversions of four noise-
free and four noise-contaminated synthetic data sets, uncertainty appraisal analysis
was performed by DE algorithm. Additionally, one field data from Italy was inverted
for the determination of applicability of DE on Rayleigh wave dispersion curves. The
inversion results using DE were found successful for accuracy and convergence speed.
Zhu et al., (2015) proposed DE algorithm for the inversion of full waveform
microseismic data. According to the inversion results of the synthetic data, accurate
solutions were achieved with less computation costs. Pan et al., (2018) carried out high
dimensional seismic inversion using DE algorithm with three subpopulations and two
phases mutation strategy. Proposed DE approach was used for the inversion of both
synthetic and field data. Faster convergence and higher quality solutions were obtained
using proposed DE algorithm for high dimensional seismic inversion. Ribeiro et al.,
(2019) used adaptive DE algorithm in seismic processing for improving parameter

48
estimations. Inversion of the synthetic and real datasets showed promising results for
robustness and convergence speed.

Li et al., (2020) proposed new DE based algorithm called modified Boltzmann


Annealing Differential Evolution algorithm for inversion of resistivity. The proposed
algorithm was designed as combination of multi population DE and annealing strategy.
The inversion results of synthetic and field examples showed that proposed algorithm
was effective for convergence speed compared to the standard DE algorithm.

Li and Yin (2012) applied DE algorithm on SP data for the estimations of six model
parameters. Synthetic without noise, noise-contaminated synthetic and field data from
Türkiye was used for the inversion study. According to the inversion results and
frequency distribution of model parameters, DE algorithm was defined efficient for SP
anomalies. Balkaya (2013) used DE algorithm for the inversion of both SP and vertical
electrical sounding (VES) anomalies. The study includes the inversion of noise-free
and noisy synthetic anomalies and field data sets for SP and VES. The inversion results
of estimated parameters and uncertainty appraisal analysis using Metropolis–Hastings
(M–H) algorithm show that DE performs fast approximation to the solution.

3.5 Applications of Other Metaheuristic Algorithms

New nature inspired metaheuristic algorithms apart from the popular metaheuristic
algorithms of GA, PSO, DE and SA have been also implemented since few decades to
the geophysical inversion studies for the model parameter estimations of various
synthetic and field anomalies. Therefore, investigation of the various metaheuristic
applications from previous published literature is helpful to understand the
developments of inversion methods with nature inspired algorithms, algorithm design
and applications for different types of geophysical anomalies.

Table 3.1 shows geophysical inversions studies using various metaheuristic


algorithms with their compared studies, which were applied in last seven years.

49
Table 3.1 Current applications of some metaheuristic algorithms for geophysical inversions between
2016 to 2022
Applied Metaheuristic Inverted Compared
Reference
Algorithm Geophysical Data Algorithm/Method
Di Maio et al., Least Squares, GA, PSO,
Genetic Price Algorithm Self-Potential
(2016) SA
Poormirzaee (2017) Bat Algorithm Seismic -
Vertical Electrical
Chandra et al. PSO, Ant Colony
Grey Wolf Algorithm Sounding,
(2017) Algorithm
Magnetotelluric
Cheng-Yu et al., Shuffled Frog-Leaping
Rayleigh PSO
(2017) Algorithm
Agarval et al., Magnetic, Gravity,
Grey Wolf Algorithm PSO
(2018) Self-Potential
Alkan and Balkaya Differential Search
Electromagnetic Least Squares
(2018) Algorithm
Alvandi and Asil Artificial Bee Colony
Gravity Least Squares
(2018) Algorithm
Magnetotelluric,
Li et al., (2018) Grey Wolf Algorithm PSO, SA
Resistivity
PSO, Singular Value
Sungkono and
Black Hole Algorithm Self-Potential Decomposition
Warnana (2018)
Algorithm, Least Squares
Vashisth et al., Whale Optimization PSO, Grey Wolf
Self-Potential
(2018) Algorithm Algorithm
Shuffled Frog-Leaping
Wang et al., (2018) Rayleigh PSO
Algorithm
Abdelazeem et al., Least Squares, Black Hole
Whale Algorithm Self-Potential
(2019) Algorithm, SA
Di Maio et al.,
Genetic Price Algorithm Self-Potential PSO, Very Fast SA
(2019)
Poormirzaee et al.,
Modified Bat Algorithm Seismic Bat Algorithm
(2019)
PSO, Least Squares,
Di Maio et al.,
Genetic Price Algorithm Magnetic Curve Matching
(2020)
Techniques

50
Table 3.1 Continues
Applied Metaheuristic Inverted Compared
Reference
Algorithm Geophysical Data Algorithm/Method
PSO, Genetic Price
Gobashy et al., Whale Optimization
Self-Potential Algorithm, Levenberg–
(2020) Algorithm
Marquardt Method
Haryono et al., Genetic Price Algorithm,
Crow Search Algorithm Self-Potential
(2020) Very Fast SA
Poormirzaee et al., Modified Flower Flower Pollination
Rayleigh
(2020) Pollination Algorithm Algorithm
Clonal Selection
Yan et al., (2020) Seismic GA, PSO, DE
Algorithm
Balkaya and Kaftan Differential Search GA, PSO, Genetic Price
Magnetic
(2021) Algorithm Algorithm, Very Fast SA
Joolaei et al., Imperialist Competitive
Gravity -
(2021) Algorithm
Lévy Gradient Descent SA, Monte Carlo Markov
Zhang et al., (2021) Gravity, Seismic
Algorithm Chain Algorithm
Social Spider
Ben et al., (2022) Magnetic PSO
Optimization Algorithm
Modified Whale Improved Whale
Liang et al., (2022) Seismic
Optimization Algorithm Optimization Algorithm
Ni and Liang Multioperator Whale Whale Optimization
Seismic
(2022) Optimization Algorithm Algorithm
Manta-Ray Foraging
Ben et al., (2022) Magnetic PSO, SA, DE
Optimization
Bouchaoui et al., Ant Colony Optimization Vertical Electrical
GA, SA
(2022) Algorithm Sounding

Kumar and Gupta (2012) proposed Harmony Search Algorithm for the inversions
of nonlinear 1D seismic waveform to minimize error between observed and synthetic
model parameters. Control parameters were found for the inversion studies with
evaluating pitch adjusting rate and band width of the of the algorithm. Harmony Search
Algorithm was successful for model parameter estimations of without noise and noise-
contaminated synthetic seismic data. Conti et al., (2013) developed an inversion study
for 3D seismic data using Ant Colony Optimization Algorithm. Without noise and
30% noise-contaminated synthetic data were used for the inversions. The inversion

51
results were also compared with the results obtained from a traditional method of
forward convolution. As a result of the inversion and comparison studies, higher
convergence speed and effectiveness were achieved with Ant Colony Optimization
Algorithm than traditional method. Liu et al., (2014) was also implemented Ant
Colony Optimization Algorithm to the inversion of potential field data to find the
model parameters of fixed and complicated shape models. At the end of the inversions
of synthetic and field data of gravity and magnetic anomalies, good agreement was
achieved with estimated and true model parameters and Ant Colony Optimization
Algorithm was offered good optimization capability than linear inversions and other
stochastic metaheuristics. Song et al., (2014a) proposed Differential Search Algorithm
for inversion of Rayleigh wave dispersion curve. Four without noise and four noise-
contaminated synthetic data were inverted for evaluation of the efficiency of the
algorithm and comparison of the results with the GA. According to the convergence
speed, robustness and accuracy, Differential Search Algorithm was found successful
compared to the GA. Shadmehri et al., (2015) used Ant Colony Algorithm for
inversion of synthetic and field gravity data. The inversion results show that Ant
Colony Algorithm was a good option to estimate the model parameters of regional
gravity field data.

It is clearly seen that the researchers have paid great attention for developing new
nature-inspired metaheuristic algorithms for geophysical inversion studies using
various synthetic and field data sets from gravity, SP, magnetic, seismic, Rayleigh,
vertical electrical sounding and MT anomalies. In addition to the most popular
metaheuristic algorithms, researchers have studied inversion of geophysical anomalies
in last five years using Genetic Price, Grey Wolf, Bat, Whale, Shuffled Frog-Leaping,
Differential Search, Black Hole, Artificial Bee, Crow Search, Imperialist Competitive,
Modified Flower Pollination, Clonal Selection and Lévy Gradient Descent Algorithms
according to the Table 2.1. These studies include comparison of the proposed
metaheuristic algorithms with other popular metaheuristic algorithms (GA, PSO, DE
and SA etc.), relatively new metaheuristic algorithms (Differential Search Algorithm,
Genetic Price Algorithm etc.) or derivative based inversion techniques such as least
squares method.

52
CHAPTER FOUR
CUCKOO SEARCH ALGORITHM FOR SELF-POTENTIAL AND
GRAVITY ANOMALIES

In nature, cuckoo birds stand out among other creatures with their unusual breeding
system based on brood parasitism. The cuckoo birds generally find a new nest that
belongs to another species for their eggs. The she-bird puts her egg into the nest of a
hosted bird and throws an egg of the hosted bird off the nest to increase her own egg’s
surviving probability. CSA was created after the inspiration of the cuckoos’ survival
strategy, and the standard CSA is described on the following idealized:

i) A cuckoo put an own egg in the nest of another kind of bird randomly, and
the cuckoo takes away an egg of hosted bird,
ii) The best eggs in the nest will survive for the next generations,
iii) The host bird may throw out the cuckoo’s egg from the nest or leave the nest
to rebuild at somewhere else. In this situation, the probability of recognition of
the cuckoo’s egg by a hosted bird is ρa ∈ (0, 1).

On the other hand, CSA is improved by using Lévy flights instead of conventional
random walks (Yang and Deb, 2013). The Lévy flights are based on random walks
with a straight flight, and sudden turn maneuvers for exploring larger scale search area
compared to ordinary random walks. Pseudo-random number generation in a Lévy
flight is defined by choosing a direction randomly and generation of steps according
to the distribution of Lévy, respectively. This means a combination of a local and
global explorative random walks for CSA and tuning the parameter ρa controls the
algorithm. These searching capabilities make CSA more efficient than the other
methods depending on the better convergence to the global optimality with a higher
probability. The random walks are given in equation 1 and 2, respectively to generate
new solution via Lévy flights after the setting the best of all parameters (Yang, 2014).

𝑥𝑖𝑡+1 = 𝑥𝑖𝑡 +∝ 𝑠 ⊗ 𝐻 (𝜌𝒶 − 𝜀 ) ⊗ (𝑥𝑗𝑡 − 𝑥𝑘𝑡 ) (4.1)

53
𝑥𝑖𝑡+1 = 𝑥𝑖𝑡 +∝ 𝐿(𝑠, 𝜆) (4.2)

where xi is an individual in the solution space, and i ranges between lower and upper
limits (i=1, 2, …, n) which n denotes as host nests. Time factor is indicated by t and
t+1 for sequential positions of instantaneous time. Therefore, 𝑥𝑖𝑡 is the current position
selected by random permutation, 𝑥𝑖𝑡+1 is the next position, H(u) is a Heaviside function,
α is scaling factor, ε is a uniformly distributed random number, s is the step size, 𝑥𝑗𝑡 and
𝑥𝑘𝑡 are two different random solutions by random permutation. Also, indicates the
entry-wise product. The Lévy distribution is formulated by L(s,λ), and it is given as
follows:

𝜆Γ(𝜆)𝑠𝑖𝑛(𝜋𝜆⁄2) 1
𝐿(𝑠, 𝜆) = ∙ , (𝑠 ≫ 𝑠0 > 0) (4.3)
𝜋 𝑠 1+𝜆

The standard gamma function is represented by Γ(λ), and it is valid for large steps
when s>0. In Eq. 3 scaling factor should be as α>0 according to the scales of the
problem.

Yang (2014) indicates that the DE, PSO, and SA are the special forms of the CSA,
and the CSA may be considered as an algorithm combining these into one single
algorithm. Thus, it is expected to be more efficient than DE, PSO, and SA (Yang,
2014). According to the rules of the CSA, a flowchart is illustrated in Figure 4.1 with
the number of population (Np) and probability (ρa) as the algorithm-based (control)
parameters for the inversion of SP anomaly.

54
Figure 4.1 Flowchart of the cuckoo search algorithm (Turan-Karaoğlan and Göktürkler, 2021)

55
Cuckoo search algorithm (CSA) by Yang and Deb (2009) uses the strategy
employed by cuckoo birds to raise their numbers in the population, is becoming more
popular among the metaheuristic algorithms. According to the recent studies (Walton
et al., 2011; Yang and Deb, 2013), CSA attracts the attention due to its simplicity and
efficiency over PSO and GA because CSA use the Lévy flights instead of isotropic
random walks (Yang, 2014). Therefore, researchers have focused on CSA for solving
several engineering problems. For instance, Ma et al. (2013) carried out a parameter
estimation study for their model of commercial photovoltaic generators using CSA.
The paper presents that CSA is successful for the lowest rms compared to GA and PSO
algorithms. In another study, Nguyen et al. (2016) investigated CSA to solve the
problem for optimal power flow of a hydrothermal system. CSA was found more
effective than PSO according to the results of the optimization of hydrothermal system
in the study. On the other hand, modified versions of the CSA are presented in several
studies for the optimization problems. For example, Mareli and Twala (2018)
developed CSA with three different modifications for optimization studies based on
dynamically increasing switching parameters and modified CSAs were validated with
mathematical test functions. Salgotra et al. (2018) proposed modified versions of the
CSA with improved exploration and exploitation properties, and it was applied on
numerical and statistical tests to compare the results with other metaheuristics.

A comprehensive literature review will yield that there are no studies applying the
CSA to SP anomalies. Therefore, the conventional CSA (Yang and Deb, 2009) was
used to invert the SP data in this paper because CSA contains only two user defined
simple control parameters for tuning compared to other optimization algorithms.
Theoretical data set by a single structure and three field data sets from Bavarian
(Germany), Süleymanköy (Türkiye), and Malachite (USA) were used to obtain the
model parameters by the CSA. Additionally, inversion results were compared with
other results obtained by various researchers employing different metaheuristics from
the current studies.

56
4.1 Inversion of SP Anomalies with Cuckoo Search Algorithm

In applied geophysics, the method of SP is carried out by measuring the electrical


fields, which naturally occurred inside the Earth, along its surface. The natural
electrical fields occur because of the mechanisms, including electrochemical processes
of the earth, thermoelectric coupling, and electro-kinetic coupling (Corwin and
Hoover, 1979). The SP, as a cost-effective geophysical method, is helpful for mineral
exploration (e.g. Meiser, 1962), landslide survey (e.g. Bogoslovsky and Ogilvy, 1977),
and groundwater investigation (e.g. Göktürkler et al., 2008). Generally, SP anomalies
are assumed to be related with the subsurface bodies having some simple geometrical
shapes including a semi-infinite vertical cylinder (simple geometry-1), an infinitely
long horizontal cylinder (simple geometry-2), and a sphere (simple geometry-3). The
model parameters of the anomaly consisting of depth, shape factor, and polarization
angle might be estimated by the inversion of SP anomalies using various interpretation
techniques such as least-square approach (e.g. El-Araby, 2004), Fourier and Hilbert
transform (e.g. Di Maio, 2017), gradient and derivative method (e.g. Abdelrahman et
al., 1997). A comprehensive literature survey shows that SP anomalies are inverted by
using a variety of the interpretation methods from the derivative-based to metaheuristic
algorithms. The derivative-based optimization algorithms use deterministic
approaches to reach a solution with a well-designed initial model. These algorithms
converge to a solution faster than a metaheuristic method, but it can encounter a local
minimum instead of the global one. On the other hand, the metaheuristic methods are
not dependent on the initial model. Therefore, a large search space can be scanned by
these methods effectively to reach a global minimum by avoiding local minima.
Hence, the metaheuristics are frequently preferred for the optimization problems
despite their higher computational cost caused by a large number of forward solutions
(Göktürkler, 2011).

4.1.1 Mathematical Model of Self-Potential Anomaly

The SP anomaly caused by a body being polarized with a defined geometrical shape
can be calculated by a mathematical model. It may be in the form of a simple geometry-

57
1 (SGM-1), a simple geometry-2 (SGM-2) or a simple geometry-3 (SGM-3). The
mathematical model for the potential caused by such a body is given in below (Yüngül,
1950; Göktürkler and Balkaya, 2012):

(𝑥−𝑥0 )𝑐𝑜𝑠𝜃+𝑧0 𝑠𝑖𝑛𝜃


𝑉(𝑥, 𝑥0 , 𝐾, 𝜃, 𝑧0 , 𝑞) = 𝐾 [(𝑥−𝑥0 )2 +𝑧0 2 ]𝑞
(4.4)

where, x [m] is the distance in horizontal along the observation line, x0 and z0 [m]
define the exact origin and depth of the causative body, respectively. The
dimensionless shape factor is represented by q, and it is taken as 0.5, 1.0 and 1.5 for
SGM-1, SGM-2 and SGM-3, respectively. K and θ are the electric dipole moment
[mVm2q-1] and the polarization angle [°], respectively. The model parameters for the
geometrical bodies mentioned above are given in Figure 4.2 (Turan-Karaoğlan and
Göktürkler, 2021).

Figure 4.2 Illustration of the primary geometrical bodies. (a) semi-infinite vertical cylinder, (b) infinitely
long horizontal cylinder and (c) sphere (Turan-Karaoğlan and Göktürkler, 2021)

4.1.2 Parameter Tuning of Self-Potential Data for Cuckoo Search Algorithm

Using the noise-free synthetic data, a parameter tuning study was carried out for the
parameters of Np and ρa to determine their best values. The results, which was obtained
by a computer having 3.20 GHz processor with a memory of 4 GB, are given in Table
4.1 with mean CPU time for each Np. Each row shows the best values in Table 4.1 for
the model parameters together with their root mean square (rms) value statistics at the
end of five different run of the CSA for the variations of the algorithm-based
parameters. Each run included 300 iterations for each pair of the algorithm-based

58
parameters Np and ρa. The rms values were calculated by the following equation for
the square root of error energy:

1 𝑜𝑏𝑠 2
Φ = ∑𝑁
𝑖=1(𝑑𝑖 − 𝑑𝑖𝑐𝑎𝑙 ) (4.5)
𝑁

Here, i denotes each observation, N is the number of the data, dobs is the observed
data, and dcal is the calculated data.

A synthetic SP anomaly caused by an SGM-2 (Figure 4.2b) was generated for


numerical tests (Figure 4.3). The anomaly was calculated along with a profile of 500
m with a 10-m interval for sampling. The model parameters were taken as K=1000
mVm2q-1, θ=170°, z0=30 m, q=1, and x0=250 m. The resulting anomaly displays an
amplitude variation in the interval of -15 and 20 mV.

Figure 4.3 Synthetic SP anomaly for an infinitely long horizontal cylinder (Turan-Karaoğlan and
Göktürkler, 2021)

Investigation of the effects of the algorithm-based parameters is carried out by the


changing in Np between 15 to 50 with five increments and ρa between 0.05 and 0.45
with 0.10 increments. According to the parameter tuning results, the lowest mean rms
was obtained as 0.145 mV when the algorithm-based parameters are taken as 45 for
Np and 0.15 for ρa. At this point, the extremum values of rms are 0.063 mV and 0.210
mV, respectively.

59
Table 4.1 Parameter tuning study for CSA with the theoretical data without noise for SP anomaly (The
best result of the parameter tuning study is shown in boldface type)

Model Parameters rms (mV) Mean


Np ρa K θ z x0 Elapsed
q Min Max Mean SD
[mVm2q-1] [°] [m] [m] Time* [s]
0.05 1450.35 171.23 32.46 1.04 250.47 0.053 0.339 0.169 0.112
0.15 760.72 169.97 28.24 0.97 250.50 0.066 0.220 0.157 0.059
15 0.25 1208.34 171.11 31.24 1.02 250.67 0.136 0.300 0.220 0.065 17.89
0.35 1115.06 170.00 31.00 1.01 250.13 0.154 0.329 0.242 0.067
0.45 816.55 170.86 27.94 0.98 250.65 0.175 0.365 0.252 0.078
0.05 980.62 168.69 29.72 1.00 249.48 0.065 0.231 0.153 0.065
0.15 1647.99 187.91 32.86 1.05 250.64 0.109 0.374 0.253 0.100
20 0.25 1124.86 171.28 29.48 1.01 250.97 0.146 0.315 0.255 0.074 22.54
0.35 1342.31 171.15 30.85 1.03 250.38 0.197 0.348 0.251 0.060
0.45 1151.03 171.94 30.05 1.02 250.88 0.111 0.429 0.239 0.120
0.05 1115.95 169.81 30.52 1.01 250.27 0.129 0.274 0.204 0.062
0.15 872.57 170.24 28.68 0.99 249.69 0.077 0.229 0.178 0.060
25 0.25 1388.16 171.23 31.74 1.04 250.40 0.185 0.248 0.208 0.028 27.30
0.35 843.22 170.74 29.40 0.98 251.15 0.190 0.380 0.250 0.076
0.45 1342.25 170.25 32.13 1.03 250.06 0.102 0.248 0.171 0.066
0.05 1488.92 170.74 32.84 1.04 251.31 0.153 0.327 0.229 0.070
0.15 1157.55 170.52 29.72 1.02 251.09 0.214 0.321 0.273 0.050
30 0.25 1257.02 170.09 30.92 1.02 250.10 0.153 0.274 0.210 0.044 31.95
0.35 1540.83 171.33 31.72 1.05 251.35 0.127 0.336 0.209 0.079
0.45 1276.22 170.03 31.63 1.02 250.23 0.166 0.334 0.228 0.069
0.05 983.08 169.70 29.71 1.00 250.06 0.071 0.231 0.154 0.065
0.15 1403.27 171.60 31.80 1.04 250.60 0.169 0.269 0.228 0.041
35 0.25 1205.82 169.54 32.29 1.02 249.96 0.127 0.294 0.218 0.061 36.59
0.35 1474.18 171.14 32.54 1.04 250.54 0.117 0.206 0.165 0.043
0.45 905.10 169.43 29.45 0.99 249.82 0.072 0.194 0.159 0.051
0.05 1500.00 169.98 32.92 1.04 249.62 0.130 0.224 0.179 0.040
0.15 1074.74 169.61 30.47 1.01 249.22 0.126 0.301 0.197 0.068
40 0.25 1198.30 169.23 30.69 1.02 249.75 0.151 0.276 0.204 0.046 41.38
0.35 1408.76 169.42 32.89 1.03 248.76 0.170 0.314 0.261 0.059
0.45 973.34 169.17 30.09 1.00 249.81 0.122 0.287 0.202 0.067
0.05 898.77 170.64 29.45 0.99 250.28 0.088 0.308 0.174 0.085
0.15 1083.61 169.87 30.60 1.01 249.78 0.063 0.210 0.145 0.055
45 0.25 791.18 168.49 28.24 0.98 249.23 0.179 0.269 0.212 0.034 45.65
0.35 866.05 170.28 28.76 0.99 249.90 0.124 0.200 0.147 0.030
0.45 890.32 170.32 30.05 0.99 249.85 0.123 0.203 0.180 0.032
0.05 1317.38 170.22 30.05 1.03 249.89 0.125 0.397 0.276 0.105
0.15 1289.02 169.89 30.87 1.03 249.92 0.158 0.278 0.223 0.053
50 0.25 1276.39 170.19 31.52 1.02 249.71 0.142 0.281 0.187 0.057 50.47
0.35 914.37 170.09 29.56 0.99 250.25 0.074 0.354 0.193 0.104
0.45 821.72 168.96 28.65 0.98 249.39 0.083 0.209 0.165 0.055
*
Average value of Mean Elapsed Time for all ρa parameters in each Np

The standard deviation was 0.055 mV. On the other hand, the model parameters for
the SP anomaly are obtained as 1083.6 mVm2q-1 for K, 169.87 m for z0, 30° for θ, 1.01
for q and 249.78 m for x0 with the best values of the algorithm-based parameters.

60
Additionally, acceptable results are obtained from the parameter tuning study for the
mean CPU time. In the rest of the present study, the tests, including the ones with the
field data sets were performed with the same values of the algorithm-based parameters
mentioned previously.

Figure 4.4 shows the plots of rms values versus the parameter Np for various values
of the parameter ρa. The deviations for all control parameters are shown with the error
bar integrations of rms values according to the related standard deviations given in
Table 4.1. Decreasing characteristic of rms with low deviations is observed in Figure
4b for 0.15 of ρa and Np between 30 and 45.

Figure 4.4 Change of the rms values with respect to Np for: (a) ρa =0.05, (b) ρa =0.15, (c) ρa =0.25, (d)
ρa =0.35 and (e) ρa =0.45.

61
4.1.3 Numerical Tests and Metropolis Hastings Algorithm for Self-Potential

Like the parameter tuning, the tests for the synthetic and field data sets were carried
out as the five independent runs, and each including 300 iterations. During the tests,
the rms values and plots of the error energy 300 iterations were considered to evaluate
the inversion results. Moreover, a Markov Chain Monte Carlo (MCMC) algorithm
with the Metropolis-Hastings (M-H) method, introduced by Metropolis et al. (1953)
and put into a generalized form by Hastings (1970) to indicate the accuracy of model
estimations for metaheuristic methods, was performed for the uncertainty appraisal
analysis for each model parameter. The M-H method was implemented by employing
a SA algorithm with a fixed temperature value, in other words without a decrease in
the temperature (Sen and Stoffa, 1996; Balkaya, 2013). Considering the field data sets,
the parameter estimations were also compared with those from the prior works in the
literature used several metaheuristic methods (GA, PSO, DE and SA) with the same
field data sets.

4.1.4 Tests with Theoretical Data for Self-Potential

Both the noise-contaminated and without noise cases were considered for inverting
the synthetic anomalies. The noise-contaminated data set was obtained by adding
pseudo-random numbers of the Gaussian distribution and zero-mean to the noise-free
data set. The pseudo-random numbers, with ±2 mV standard deviation, were generated
by using one of the algorithms given in Galassi et al. 2009. Table 4.2 gives the
solutions resulted from the CSA, together with the true values and the bounds set for
each model parameter in the search space.

Table 4.2 Solutions based on the theoretical data sets of SP anomaly by CSA.
True Parameter Bounds CSA Estimations
Parameters
values Minimum Maximum Noise-free data Noisy data
K [mVm2q-1] 1000 -1500 1500 1083.61 1077.09
𝜃 [°] 170 0 200 169.87 170.61
z0 [m] 30 0.1 50 30.60 31.00
q 1 0.1 2.0 1.01 1.01
x0 [m] 250 50 300 249.78 250.41
rms [mV] - - - 0.063 2.042

62
The results from noise-free and noise-contaminated data sets are in the proximity
of the true model parameters. The solution to the noise-contaminated data yielded a
rms value of 2.042 mV, which is around the noise level of ±2 mV. Besides, the results
from the data without noise are better than those from the noisy data naturally.

Figure 4.5 illustrates the comparisons for observed and calculated anomalies for
noise-free (Figure 4.5b) and noise-contaminated (Figure 4.5d) data with the plots of
the error energies for theoretical data without noise (Figure 4.5a) and with noise
(Figure 4.5c). Good convergence characteristics of the solutions are achieved in both
theoretical data. Additionally, change of the model parameters are presented in Figure
4.6 with respect to the iteration number.

Figure 4.5 Change of the error energy during the solutions based on the theoretical data (a) without
noise and (c) with noise for 300 iterations, and comparison of observed and calculated anomaly results
for (b) noise-free and (d) noisy SP data (Turan-Karaoğlan and Göktürkler, 2021)

63
Figure 4.6 Change of the model parameters (a-e) noise-free and (f-j) noisy theoretical data sets
according to iteration number (Turan-Karaoğlan and Göktürkler, 2021)

64
Figure 4.7 Error energy maps of SP synthetic noise-free data for model parameters using CSA

65
Error energy topography maps are used to understand the effects of the selected
model parameter couples calculated by the inversions according to the corresponding
ranges of the search space (Ekinci et al., 2021). Relations of the model parameter
couples of the noise free SP anomaly are shown in the Fig. 4.7 for each combination
of the model parameter which are observed by the inversions using CSA. Scale of the
error energy is located at the right side of the plots. True values of the noise free SP
anomaly are given in the error energy maps with black dots. On the other hand,
estimated model parameter values are shown in the maps with red dots. Relations of
the error energies of the model parameter couples occur closed shape contours.

The results of an uncertainty appraisal analysis for each model parameter are given
in Figure 4.8 The analysis is based on the M-H method with 90001 samples for each
model parameter. Based on the histograms resulted from the M-H method, the CSA
generates the solutions in the vicinity of the true parameters with acceptable accuracy
for both data sets. Additionally, estimations, mean values and standard deviations by
M-H algorithm for the model parameters are also given in Table 4.3.

Figure 4.8 Histograms via the M-H algorithm for (a-e) the noise-free and (f-j) noisy data sets. The
parameters obtained by the CSA are also indicated on each histogram (Turan-Karaoğlan and Göktürkler,
2021)

66
Table 4.3 CSA estimations, mean values and standard deviations (SD) for the synthetic data obtained
from M-H algorithm
Noise-free data Noisy data
Parameters CSA Mean CSA Mean
SD SD
Estimation Value Estimation Value
K [mVm2q-1] 1083.61 1049.80 64.312 1077.09 1206.01 98.033
𝜃 [°] 169.87 170.03 4.412 170.61 170.08 4.450
z0 [m] 30.60 30.51 1.964 31.00 30.76 2.652
q 1.01 1.01 0.011 1.01 1.02 0.018
x0 [m] 249.78 249.71 4.178 250.41 250.14 4.282

Some large differences in CSA estimations and mean values are present in the
parameter K for both without noise and noise-contaminated data according to the
results from the M-H algorithm. The highest standard deviations are obtained as
64.312 and 98.033 for without noise and noise-contaminated data, respectively. On the
other hand, the standard deviation values are 4.412, 1.964, 0.011 and 4.178 for noise-
free data, 4.450, 2.652, 0.018 and 4.282 of the model parameters of 𝜃, z0, q and x0. The
minimum standard deviations are obtained for the model parameter q for both without
noise and noise-contaminated data sets.

In addition to without noise and noise-contaminated synthetic data, without noise


multiple synthetic SP data are used for the inversion using CSA. The multiple synthetic
SP anomaly includes two separate anomalies called Source-1 and Source-2. The results
of error energy and SP anomaly (observed and calculated) are shown in Fig. 4.9 for
600 iteration using CSA.

Figure 4.9 Error energy (a) and inversion result of multiple synthetic data for 600 iterations

67
68
Figure 4.10 Change of the model parameters (a-e) Source-1 and (f-j) Source-2 for multiple synthetic data sets according to iteration number.
For multiple SP synthetic anomaly, true values of the anomaly (both for Source-1
and Source-2), parameter bounds of Source-1 and Source-2 and model parameter
estimations by CSA are listed in Table 4.4 for the parameters of the SP anomaly. The
results show that estimated model parameters are close to the true values with 1.76 mV
rms value. On the other hand, values of the model parameters based on the iteration
are shown in the Fig. 4.10 for both Source-1 and Source-2.

Table 4.4 True values, parameter bounds and CSA estimations for the multiple synthetic data
Bounds for Bounds for
True Values CSA Estimations
Parameters Source-1 Source-2
Source-1 Source-2 Min Max Min Max Source-1 Source-2
K [mVm2q-1] -1500 115 -2000 1000 50 500 -1020.46 112.63
𝜃 [°] 165 235 0 180 90 360 163.050 235.12
z0 [m] 5 10 0 15 5 25 4.05 10.15
q 1 0.5 0 2 0 2 0.94 0.49
x0 [m] 75 275 0 200 100 500 73.79 274.90
rms [mV] - - - - - - 1.76

4.1.5 Tests with Field Data for Self-Potential

The SP anomalies from Bavarian woods in Germany, Süleymanköy in Türkiye, and


Malachite mine in Colorado (USA) were considered for the field data examples to
investigate the applicability of the CSA. These field data sets are preferred for the
inversion because they are well-known and well-studied benchmark field data sets in
the literature.

4.1.5.1 Bavarian Woods Anomaly (Germany)

The field data obtained by Meiser (1962) over a graphite deposit has both positive
and negative SP values along 520 m profile, and the maximum amplitude is observed
at 250 m of the profile with a value of approximately -500 mV. The search spaces were
set as K ∈ [5000, 40000], θ ∈ [-80, -40], z0 ∈ [30, 60], q ∈ [0.1, 2.0], and x0 ∈ [240,
280]. Figure 4.11 illustrates the results by the CSA for the Bavarian anomaly.

69
Figure 4.11 The results from the Bavarian woods, Germany SP data set. (a) change of the error energy
and (b) the comparison of the observed and calculated anomalies (Turan-Karaoglan and Gokturkler,
2021)

The estimations are compared with the previous studies in the literature using
different metaheuristic algorithms. This comparison is given in Table 4.5. According
to this, the CSA yields very similar results to those obtained from the other methods.
In addition, it is evident that the solutions from the CSA and PSO are in good
accordance.

Table 4.5 The solution by CSA in comparison with the prior studies (The Bavarian woods anomaly)
Göktürkler and Balkaya (2012) Di Maio Present
Parameters et al. (2016) Study
GA PSO SA Genetic-Price CSA
K [mVm2q-1] 21272.91 33343.87 26257.42 39951.67 36636.86
𝜃 [°] -51.29 -48.60 -49.98 -47.30 -48.15
z0 [m] 45.03 47.59 45.99 49.50 48.31
q 0.97 1.02 0.99 1.0 1.03
x0 [m] 268.79 269.88 269.17 271.10 270.10
rms [mV] 11.40 10.95 11.09 - 10.95

4.1.5.2 Süleymanköy Anomaly (Türkiye)

The Süleymanköy anomaly by Yüngül (1950) over metallic sulfide deposits


displays a change between a minimum value of -225 mV and a maximum value of 100
mV along with a profile of 260 m. The search spaces were set as K ∈ [-60000, 100000],
θ ∈ [5, 50], z0 ∈ [20, 50], q ∈ [0.5, 1.5] and x0 ∈ [50, 100]. The results by the CSA are

70
given Figure 4.12 A comparison of the solution from the CSA with the ones from the
studies based on the different metaheuristics is given in Table 4.6. The CSA produces
results, which are like the ones from the other metaheuristics. Again, the similarity in
terms of the yielded solutions among the CSA, GA, PSO, and SA is observed here.

Figure 4.12 The results from the Süleymanköy (Türkiye) SP data set. (a) change of the error energy and
(b) the comparison of the field anomaly and model response (Turan-Karaoğlan and Göktürkler, 2021)

Table 4.6. The solution by CSA in comparison with the prior studies (The Süleymanköy anomaly)
Tlas and
Biswas Present
Asfahani Göktürkler and Balkaya (2012)
Parameters (2017) Study
(2008)
ASA* VFSA** GA PSO SA CSA
2q-1
K [mVm ] -904.03 10079.9 -36724.18 -40549.97 -44771.77 -40465.40
𝜃 [°] 17.76 72.2 22.85 22.44 22.21 22.43
z0 [m] 35.41 25.4 32.68 33.11 33.59 33.14
q 1.19 1.0 1.16 1.17 1.18 1.17
x0 [m] 15.47 72.2 79.24 79.13 79.09 79.13
rms [mV] - - 12.20 12.19 12.19 12.19
*
Adaptive simulated annealing **very fast simulated annealing

4.1.5.3 Malachite Mine Anomaly (USA)

The field anomaly of Malachite mine in Colorado (USA) was observed by Heiland
et al. (1945). The anomaly along a line of 200 m is characterized by the negative values
having a maximum of -225 mV. The search spaces were chosen as K ∈ [-1500, 100],
θ ∈ [50, 120], z0 ∈ [0.1, 50], q ∈ [0.1, 2.0] and x0 ∈ [1.0, 184]. The results are given in
Figure 4.13 and Table 4.7. The similarity of the solutions yielded by the CSA, PSO,
and DE is noticeable in Table 4.7.

71
Figure 4.13 The results from the Malachite mine (Colorado, USA) SP data set. (a) change of the error
energy and (b) the comparison of the observed and calculated anomalies (Turan-Karaoğlan and
Göktürkler, 2021)

Table 4.7 The solution by CSA in comparison with the prior studies (The Malachite mine anomaly)
Biswas and
Present
Sharma Balkaya (2013)
Parameters Study
(2015)
VFSA* DE PSO CSA
2q-1
K [mVm ] -224.4 -436.11 -460.56 -492.34
𝜃 [°] 80.07 83.56 83.32 83.44
z0 [m] 15.2 19.21 19.51 19.79
q 0.5 0.62 0.62 0.64
x0 [m] -0.96 92.32 92.24 92.19
rms [mV] - 6.56 6.57 6.62
*
Very Fast Simulated Annealing

4.1.5.4 KTB Borehole Anomaly (Germany)

KTB Borehole Anomaly was introduced by Franke (1989) from two boreholes
which were drilled for German Continental Deep Drilling Program (Kontinentales
Tiefbohrprogramm der Bundesrepublik – KTB). Two boreholes were drilled through
the zone of Erbendorf-Vohenstrauß in Oberpfalz located in Northeast Bavaria
(Germany) for the KTB program (Biswas, 2017). Two boreholes cause two peak
negative anomaly zones in the subsurface structure (Biswas and Sharma, 2017). First
borehole was called Source-1 and its depth are about 4 km. Second borehole was called
Source-2 and it was located 200 m further from the first anomaly. Second anomaly has
also 9.1 km depth. KTB Borehole Anomaly has two negative peaks as -500 mV and -

72
600 mV magnitudes for the first and second anomalies respectively (Gobashy et al.,
2020).

Inversion results of the KTB Borehole Anomaly are achieved by CSA using two
different SP models. Firstly, the results are obtained according to the SP vertical
cylinder model for both anomalies (Source-1 and Source-2) of the KTB Borehole
Anomaly as Case-1. Table 4.8 shows parameter bounds and estimated values by CSA
of KTB Borehole Anomaly of Source-1 and Source-2 for Case-1. Additionally, the
results of error energy according to the 600 iteration and comparison of the observed
and calculated anomalies are presented in Figure 4.14 for Case-1. The results are
compared with prior studies for Case-1 in Table 9.

Figure 4.14 The results from the KTB Borehole (Germany) SP data set using vertical cylinder model
(Case-1). (a) change of the error energy and (b) the comparison of the observed and calculated anomalies

Table 4.8. The inversion results by CSA (KTB Borehole Anomaly) using vertical cylinder SP model
(Case-1)
Bounds for Bounds for
CSA Estimations
Parameters Source-1 Source-2
Min Max Min Max Source -1 Source-2
K [mVm2q-1] 100 1000 100 6000 2431.34 1555.34
𝜃 [°] -100 180 -100 180 -94.32 -83.24
z0 [m] 0 200 0 200 85.39 122.02
q 0 2 0 2 0.71 0.62
x0 [m] 500 1800 500 1800 564.16 1220.37
rms [mV] - - - - 34.26

73
Table 4.9 The solution by CSA for Case-1 in comparison with the prior studies (The KTB Borehole
anomaly)
Source Di Maio et Sungkono (2020) Present
Parameters al. (2019) Study
GPA MDE* µJADE** CSA
Source-1 K [mVm2q-1] 4669 902.12 4078.24 2431.34
𝜃 [°] 96 -99.34 -98.37 -94.32
z0 [m] 145 96.85 167.56 85.39
q 0.75 0.60 0.73 0.71
x0 [m] 1305 1294.88 1308.41 564.16
Source-2 K [mVm2q-1] 1664 395.43 2194.95 1555.34
𝜃 [°] 79 -80.09 -80.54 -83.24
z0 [m] 111 76.73 96.41 122.02
q 0.64 0.49 0.67 0.62
x0 [m] 1967 1975.5 1969.38 1220.37
*
Micro-differential evolution (MDE) variant including vectorized random mutation factor (MVDE)
**
Adaptive Micro-differential evolution MDE (µADE)

On the other hand, KTB Borehole Anomaly has been studied by many researchers
for an inversion of a field example. As a SP anomaly, KTB Borehole Anomaly is
modeled as a multiple inclined sheet geometry as Case-2 instead of defined
geometrical shapes such as SGM-1, SGM-2 or SGM-3 which are considered as SP
model in synthetic and other field examples in this thesis. Inclined sheet geometry of
the SP anomaly is illustrated in Fig. 4.15.

Figure 4.15 Illustration of the inclined sheet geometry (Modified from Murty Satyanarayana and
Haricharan, 1985)

According to the inclined sheet geometry in Case-2, the anomaly is considered as a


sheet with 2a [m] length under the subsurface in h [m] depth with 𝜃 [°] angle. SP of
the inclined sheet of the anomaly is represented by U0 [mV]. On the other hand, the

74
location of the middle point of the sheet along the X axis is represented by x0 [m]
(Sungkono, 2020). Representation of the inclined sheet SP model according to the
distance of x in the X axis of the anomaly is given in below.

{(𝑥−𝑥 )−𝑎𝑐𝑜𝑠𝜃}2 +(ℎ−𝑎𝑠𝑖𝑛𝜃)2


𝑈(𝑥, 𝑥0 , 𝑈0 , ℎ, 𝜃, 𝑎) = 𝑈0 𝑙𝑛 [{(𝑥−𝑥0 )+𝑎𝑐𝑜𝑠𝜃}2 ] (4.6)
0 +(ℎ+𝑎𝑠𝑖𝑛𝜃)2

The results of error energy according to the 600 iteration and comparison of the
observed and calculated anomalies are presented in Figure 4.16 using inclined sheet
SP model for Case-2. Additionally, model parameter estimation results and
comparisons with prior studies are listen in Table 4.10 and 4.11 for Case-2.

Figure 4.16 The results of CSA from the KTB Borehole (Germany) SP data set using inclined sheet
model (Case-2). (a) change of the error energy and (b) the comparison of the field and calculated data

Table 4.10 The inversion results by CSA (KTB Borehole Anomaly) using inclined sheet SP model
(Case-2)
Bounds for Bounds for
CSA Estimations
Parameters Source-1 Source-2
Min Max Min Max Source -1 Source-2
U0 [mV] -1000 -10 -1000 -10 -70.21 -80.72
𝜃 [°] 0 180 0 180 82.41 118.35
h [m] 10 1000 10 1000 568.55 220.27
a [m] 1 2000 1 2000 532.37 256.37
x0 [m] 0 1800 0 1800 542.35 1367.56
rms [mV] - - - - 37.06

75
Table 4.11 The solution by CSA for Case-2 in comparison with the prior studies (The KTB Borehole
anomaly)
Source Gobashy Sungkono (2020) Present
Parameters et al. (2020) Study
WOA* MVDE** µJADE*** CSA
Source-1 U0 [mV] 58.63 74.98 67.08 -70.21
𝜃 [°] 40.22 79.11 80.00 82.41
h [m] 468.53 447.09 556.99 568.55
a [m] 739.25 429.13 530.23 532.37
x0 [m] 632.70 505.52 500.00 542.35
Source-2 U0 [mV] 54.53 120.06 91.67 -80.72
𝜃 [°] 153.76 103.95 117.76 118.35
h [m] 308.21 134.82 154.15 220.27
a [m] 655.92 128.86 149.84 256.37
x0 [m] 1173.56 1224.99 1271.08 1367.56
* Whale optimization algorithm
** Micro-differential evolution (MDE) variant including vectorized random mutation factor (MVDE)
*** Adaptive Micro-differential evolution MDE (µADE)

According to the results of SP inclined sheet model from the inversion results of
KTB Borehole Anomaly using CSA, subsurface structure of the multiple anomalies of
KTB Borehole is illustrated in Fig. 4.17 under the graphic of observed (red) and
calculated (blue) anomalies including related model parameters such as h, 𝜃 and a
which are represented in Table 4.10 for Source-1 and Source-2 of the KTB Borehole
Anomaly.

76
Figure 4.17 Subsurface structure for multiple inclined sheet model of the KTB Borehole (Germany)
according to the calculated anomalies by CSA

4.1.5.5 Uncertainty Analysis of Field Data Sets

The results of the uncertainty appraisal analysis by M-H method for the field data
sets are illustrated in Figure 4.18. It can be said that the CSA has a successful
performance for the estimating of the model parameters from SP anomalies. Moreover,
Table 4.9 shows the estimation and mean values together with the standard deviations
(SD) for the model parameters for Bavarian, Süleymanköy and Malachite anomalies
obtained from the M-H algorithm with 90001 samples for each model parameter.

77
Table 4.12 CSA estimations together with mean values and standard deviations (SD) for field data sets
obtained from M-H algorithm.
Bavarian Süleymanköy Malachite
Parameters CSA Mean CSA Mean CSA Mean
SD SD SD
Estimation Value Estimation Value Estimation Value
K [mVm2q-1] 36636.86 36077.14 260.602 -40465.40 -40661.03 573.875 -492.34 -457.23 20.194
𝜃 [°] -48.15 -48.14 0.324 22.43 22.54 0.479 83.44 83.44 0.381
z0 [m] 48.31 48.15 0.315 33.14 33.26 0.331 19.79 19.51 0.411
q 1.03 1.02 0.005 1.17 1.17 0.001 0.64 0.63 0.006
x0 [m] 270.10 270.04 0.319 79.13 79.25 0.345 92.19 92.31 0.408

Figure 4.18 Histograms via the M-H algorithm for the field data sets. (a-e) The Bavarian, (f-j)
Süleymanköy and (k-o) Malachite mine anomalies. The estimated parameters by the CSA are also
indicated on each histogram. (Turan-Karaoğlan and Göktürkler, 2021)

According to the results from the M-H algorithm, the standard deviations in the
parameter K are higher than the other model parameters from the field data sets. The
highest standard deviation of the parameter K is 573.875 for the Süleymanköy data set.
On the other hand, the standard deviation values for the model parameters 𝜃, z0, q and

78
x0 are 0.324, 0.315, 0.005, 0.319 for the Bavarian; 0.479, 0.331, 0.001, 0.345 for the
Süleymanköy; and 0.381, 0.411, 0.006, 0.408 for the Malachite mine field data sets.
The minimum standard deviations are obtained for the model parameter q for the
Süleymanköy field data set.

4.2 Inversion of Gravity Anomalies with Cuckoo Search Algorithm

Gravity method, which is based on the measurement of changes in the Earth's


gravitational field, is one of the oldest methods in geophysics. There are various
studies on the gravity method in the literature. Derivative-based methods are
frequently used for the model parameter estimation of gravity anomalies. On the other
hand, metaheuristic methods are also preferred in current studies for parameter
estimation due to their advantages. Metaheuristic methods are used for the inversion
of gravity data because of they can reach the global minimum without a good initial
model and derivative calculation. Additionally, metaheuristic methods are capable for
large scale searching without getting trapped to local minimums. Literature survey
shows that CSA algorithm has not been applied on the inversion of the gravity
anomalies. Therefore, inversion of the model parameter estimation for without noise
and noise-contaminated synthetic and field (Camaguey-Cuba and Quebec-Canada)
gravity anomalies are performed using CSA algorithm.

4.2.1 Mathematical Model of Gravity Anomaly

The mathematical model of the gravity caused by a body such as SGM-1, SGM-2
and SGM-3 are given in below (Abdelrahman et al., 1989; Ekinci et al., 2016, Turan-
Karaoğlan and Göktürkler, 2022):

𝜂
𝑧0
𝑔(𝑥, 𝑥0 , 𝑧0 , 𝜂, 𝑞, 𝐴) = 𝐴 (4.7)
[(𝑥−𝑥0 )2 +𝑧02 ]𝑞

where A [mGal⋅m2q-η] is amplitude coefficient, x0 and z0 [m] define the exact origin
and depth of the causative body respectively, q and η shape factors. Physical models
and theoretical anomalies for SGM-1, SGM-2 and SGM-3 are shown in Figure 4.19.

79
Figure 4.19 The anomalies of various bodies as SGM-1, SGM-2, and SGM-3 with their physical models
(Turan-Karaoğlan and Göktürkler, 2022)

4.2.2 Parameter Tuning of Gravity Data for Cuckoo Search Anomaly

Determination of the control parameters of the algorithm is very important for the
performance of the CSA algorithm. For this reason, parameter tuning study has been
performed for the determination of the control parameters (Np and ρa) of the algorithm.
The model parameters and the rms values of the parameters are obtained with 10
independent run and they are shown in Table 4.13. Each run is performed for 300
iterations and parameter pairs of the population number (Np) and probability value (ρa)
are determined as a result of these runs. Rms values are calculated with the error energy
equation (Eq. 4.5).

After the determination of the control parameters for the inversion of gravity data,
it is aimed to determine the model parameters of the theoretical data without noise and
noise added, and the field gravity anomalies. The control parameter pairs of Np and ρa

80
are tested for suggested range of 10-50 and 0.05-0.45 respectively, to determine the
control parameters of the algorithm. The effects of the changes in the population
number (Np) and the probability value (ρa) on the results are also shown in Table 4.13.

Table 4.13 Parameter tuning study for CSA with the theoretical data without noise for gravity anomaly
(The best result of the parameter tuning study is shown in boldface type)
Model Parameters rms (mGal)
ρa Np A z0 x0
q η Min Max Mean SD
[mGal⋅m2q- η] [m] [m]
10 665.92 10.01 1.50 0.87 40.00 0.001 0.003 0.001 0.0008
20 655.09 9.95 1.49 0.86 40.00 0.001 0.003 0.002 0.0005
0.05 30 625.18 10.01 1.50 0.91 40.00 0.001 0.003 0.002 0.0004
40 503.63 10.01 1.51 1.03 39.99 0.001 0.002 0.002 0.0003
50 517.10 9.98 1.49 0.97 39.99 0.001 0.002 0.002 0.0004
10 413.53 9.98 1.49 1.07 39.99 0.002 0.006 0.003 0.0010
20 657.84 10.01 1.50 0.72 40.00 0.002 0.006 0.004 0.0012
0.15 30 592.45 9.98 1.49 0.91 39.99 0.002 0.006 0.003 0.0009
40 697.80 10.02 1.50 0.86 39.99 0.002 0.005 0.003 0.0010
50 546.95 10.00 1.50 0.96 39.99 0.001 0.005 0.003 0.0010
10 700.27 9.96 1.49 0.83 39.99 0.003 0.008 0.005 0.0017
20 643.35 10.01 1.49 0.88 40.00 0.002 0.008 0.005 0.0016
0.25 30 805.90 10.09 1.51 0.83 40.00 0.003 0.007 0.004 0.0012
40 885.28 9.99 1.50 0.75 40.00 0.001 0.007 0.004 0.0020
50 820.90 10.02 1.50 0.79 39.99 0.001 0.005 0.003 0.0011
10 672.51 9.99 1.50 0.87 39.98 0.005 0.014 0.008 0.0026
20 900.34 10.02 1.50 0.75 39.99 0.002 0.011 0.006 0.0027
0.35 30 777.83 10.01 1.50 0.82 40.00 0.003 0.009 0.005 0.0023
40 675.65 9.94 1.49 0.85 40.00 0.003 0.009 0.005 0.0019
50 376.56 9.99 1.50 1.12 40.00 0.002 0.007 0.006 0.0015
10 531.81 10.01 1.51 1.00 40.01 0.007 0.016 0.011 0.0030
20 911.12 10.02 1.51 0.76 39.99 0.003 0.012 0.009 0.0024
0.45 30 815.47 10.10 1.51 0.83 40.00 0.003 0.009 0.007 0.0020
40 366.92 10.00 1.50 1.14 40.00 0.002 0.011 0.006 0.0024
50 506.10 10.06 1.51 1.02 39.99 0.002 0.007 0.005 0.0013

According to the results of the parameter tuning studies, the lowest rms and the
standard deviation value are obtained as 0.001 mGal, and 0.0003 mGal respectively,
for the control parameters ρa is 0.05 and the Np is 40. Model parameters for these
probability value and population number, which is yielded lowest standard deviation
are obtained as 503.63 [mGal⋅m2q-η] for A, 10.01 [m] for z0, 1.51 for q, 1.03 for η and
39.99 for x0 [m]. The control parameter pairs, in which the lowest rms and standard
deviation values were obtained, are used in the inversion of the noise-contaminated
theoretical data and field data.

81
4.2.3 Tests with Theoretical Data for Gravity

Theoretical data generated with the Eq. 4.6 and noise-contaminated data sets, which
are obtained by adding a zero mean and a Gaussian distribution pseudo random
number with a standard deviation of ±0.25 mGal into the theoretical data are used for
the inversion of the without noise and noise-contaminated theoretical data sets of
gravity. The parameter estimation values, true values and search space bounds of the
model parameters are given in Table 4.14.

Table 4.14 Solutions based on the theoretical data sets of gravity anomaly by CSA.
True Parameter Bounds CSA Estimations
Parameters
Values Minimum Maximum Noise-free Data Noisy Data
A [mGal⋅m2q-η] 500 1 1000 503.63 527.67
z0 [m] 10 1 40 10.01 10.84
q 1.5 0 2 1.51 1.61
η 1 0 2 1.03 1.26
x0 [m] 40 0 80 39.99 40.06
rms [mGal] - - - 0.0011 0.256

Inversion results of the noise-free theoretical data for 300 iterations with CSA and
error energy result according to the generation number are shown in the Figure 4.20a
and 4.20b, respectively. The rms value is found as 0.0011 mGal for the theoretical data
without noise.

Figure 4.20 Inversion result of the noise-free data for 300 iteration (a) and error energy according to the
generation number (b) with CSA. (Turan-Karaoğlan and Göktürkler, 2022)

82
Figure 4.21 Convergence of the model parameters for noise-free theoretical data according to the
generation number (a-e) (Turan-Karaoğlan and Göktürkler, 2022)

Convergence of the model parameters with CSA according to the generation


number are given in Figure 4.21 for noise-free theoretical data. Model parameter
estimation values are shown in the y axis of the graphs. Additionally, the true values
of the model parameters are shown on the graphs with a dashed line.

On the other hand, noise-contaminated theoretical data is generated by the addition


of ±0.25 mGal noise into the theoretical data. Inversion results of the noise-
contaminated theoretical data for 300 iterations with CSA and error energy result
according to the generation number are shown in the Figure 4.22a and 4.22b,
respectively. Therefore, the rms value is found as 0.256 mGal for the theoretical data
with noise. As a result of the model parameter estimation study of the noise-
contaminated theoretical data with CSA, model parameters are obtained as 527.67
[mGal⋅m2q-η] for A, 10.84 [m] for z0, 1.61 for q, 1.26 for η and 40.06 for x0 [m].

Convergence of the model parameters with CSA according to the generation


number are given in Figure 4.23 for noise-contaminated theoretical data. The results
of the model parameter estimation values are shown in the y axis of the graphs.
Additionally, the true values of the model parameters are shown on the graphs with a
dashed line.

83
Figure 4.22 Inversion result of the noisy data for 300 iteration (a) and error energy according to the
generation number (b) with CSA. (Turan-Karaoğlan and Göktürkler, 2022)

Figure 4.23 Convergence of the model parameters for noisy theoretical data according to the generation
number (a-e). (Turan-Karaoğlan and Göktürkler, 2022)

Fig. 4.24 shows the error energy maps for the without noise gravity anomaly
obtained by CSA. True values of the anomaly and estimated model parameters are also
shown in the figure with black and red dots respectively. Estimated parameters, mean
values and standard deviations of the model parameters for 10 independent runs are
given Table 4.15 according to the inversion of without noise and noise-contaminated
theoretical data. The lowest standard deviation value is found as 0.11 for the η
parameter of the noise-contaminated theoretical data. Additionally, very low standard
deviation values are obtained for the q and η parameters of both without noise and
noise-contaminated theoretical data.

84
Figure 4.24 Error energy maps of gravity synthetic noise-free data for model parameters using CSA

85
Therefore, it is seen that the q and η parameters are relatively low parameter value
in all results and the average value of the results are close to the true values. On the
other hand, other parameters of A, z0 and x0, that have relatively high parameter values,
are yielded higher standard deviation results than q and η parameters. The highest
standard deviation value for parameter A is found as 176.01 because of the inversion
of the theoretical data without noise by the CSA.

Table 4.15 CSA estimation, mean and standard deviation values of noise-free and noisy theoretical data
Noise-free Data Noisy Data
Parameters CSA Mean CSA Mean
SD SD
Estimation Value Estimation Value
A [mGal⋅m2q-η] 503.63 654.9345 176.01 527.67 592.2414 93.60
z0 [m] 10.01 9.6604 1.24 10.84 9.5532 1.15
q 1.51 1.5141 0.13 1.61 1.5307 0.16
η 1.03 1.0147 0.29 1.26 0.9898 0.11
x0 [m] 39.99 39.0897 3.70 40.06 39.5108 3.24

The histograms of probability densities of the model parameters obtained from the
uncertainty analysis of M-H sampling are shown in Figure 4.25 for the without noise
and noise-contaminated theoretical data. M-H uncertainty analysis is performed using
90001 samples for each estimated model parameter. Depending on the distributions of
the model parameters in the histograms, it is seen that the CSA provides an acceptable
distribution for both noise-free and noisy theoretical data.

Figure 4.25 Histograms of relative frequency of noise-free and noisy gravity theoretical data by M-H
algorithm. (Turan-Karaoğlan and Göktürkler, 2022)
Şekil

86
4.2.4 Tests with Field Data for Gravity

4.2.4.1 Camaguey Anomaly (Cuba)

In this study, a field data from a chromite deposit in Camaguey, Cuba is used as a
first field anomaly (Davis et al., 1957). Search space boundaries for the model
parameters and the estimation result obtained from the field inversion study by CSA
are given in Table 4.16 for Camaguey field data. Model parameters of Camaguey field
data set are found as 1430.30 [mGal⋅m2q-η] for amplitude coefficient (A), 23.30 [m] for
the depth of the structure (z0), 58.72 [m] for the projection of the structure on the
surface (x0). The shape factors of the model are estimated as 1.49 and 0.18 for q and
η, respectively. The rms value for the estimated model parameters from the inversion
is calculated 0.00431 mGal with CSA. The accuracy of the estimated model parameter
values is tested with the probability density function, which is calculated for each
parameter value of the model. The results show that all estimated model parameters
are in the range of reliable boundaries.

Table 4.16 Parameter estimation results of CSA and search space boundaries of Camaguey field data
Search
CSA
Field Parameters Space
Estimations
Bounds
A [mGal⋅m2q-η] 1 – 10000 1430.30
z0 [m] 1 – 100 23.30
Camaguey
q 0.5 – 1.5 1.49
Anomaly
η 0–1 0.18
x0 [m] 0 – 120 58.72

Inversion results of the Camaguey (Cuba) field data for 300 iterations with CSA are
shown in the Figure 4.26a for estimated model parameters. In Figure 4.26b, error
energy change according to the generation number is given for the Camaguey field
data. On the other hand, convergence of the model parameters with CSA according to
the generation number are given in Figure 4.27 for Camaguey field data. The results
of the model parameter estimation values are shown in the y axis of the graphs and the
true values of the model parameters are shown on the graphs with a dashed line.

87
Figure 4.26 Inversion result of the Camaguey field anomaly for 300 iteration (a) and error energy
according to the generation number (b) with CSA. (Turan-Karaoğlan and Göktürkler, 2022)

Şekil 4.27 Convergence of the model parameters for Camaguey field anomaly according to the
generation number (a-e) (Turan-Karaoğlan and Göktürkler, 2022)

4.2.4.2 Quebec Anomaly (Canada)

Another field data used as a gravity anomaly are obtained from a sulphide ore
located in the Quebec region in Canada (Grant and West, 1965; Ekinci et al., 2016).
Search space boundaries of the model parameters and the estimation results obtained
from the inversion study by CSA are given in Table 4.17 for Quebec field data. Model
parameters of Quebec field data set are found as 204.04 [mGal⋅m2q-η] for amplitude
coefficient (A), 40.32 [m] for the depth of the structure (z0), 113.66 [m] for the
projection of the structure on the surface (x0). The shape factors of the model are
estimated as 0.92 and 0.23 for q and η, respectively. The rms value for the estimated
model parameters from the inversion is calculated as 0.0334 mGal with CSA. The

88
accuracy of the estimated model parameter values is tested with the probability density
function, which is calculated for each parameter value of the model. The results show
that all estimated model parameters are in the range of reliable boundaries for
Camaguey field data.

Table 4.17 Parameter estimation results of CSA and search space boundaries of Quebec field data
Search
CSA
Field Parameters Space
Estimations
Bounds
A [mGal⋅m2q-η] 1 – 10000 204.04
z0 [m] 1 – 100 40.32
Quebec
q 0.5 – 1.5 0.92
Anomaly
η 0–1 0.23
x0 [m] 0 – 240 113.66

Inversion results of the Quebec (Canada) field data for 300 iterations with CSA are
shown in the Figure 4.28a for observed and calculated values with estimated model
parameters. On the other hand, Figure 4.28b shows error energy change according to
the generation number is given for the Quebec field data.

Convergence of the model parameters with CSA according to the generation


number are given in Figure 4.29 for Quebec field data. The results of the model
parameter estimation values are shown in the y axis of the graphs and the true values
of the model parameters are shown on the graphs with a dashed line.

Figure 4.28 Inversion result of the Quebec field anomaly for 300 iteration (a) and error energy according
to the generation number (b) with CSA. (Turan-Karaoğlan and Göktürkler, 2022)

89
Figure 4.29 Convergence of the model parameters for Quebec field anomaly according to the generation
number (a-e). (Turan-Karaoğlan and Göktürkler, 2022)

The histograms of probability densities of the model parameters obtained from the
uncertainty analysis of M-H sampling are shown in Figure 4.30 for the Camaguey and
the Quebec field data. M-H uncertainty analysis is performed using 90001 samples for
estimated model parameters. Depending on the distributions of the model parameters
in the histograms, it is seen that the CSA provides an acceptable distribution for both
Camaguey and Quebec field data.

Şe
Figure 4.30 Histograms of relative frequency of Camaguey and Quebec field data by M-H algorithm.
(Turan-Karaoğlan and Göktürkler, 2022)

90
The estimated parameters, mean values and standard deviations of the model
parameters for 10 independent runs are given Table 4.18 according to the inversion of
Camaguey and Quebec field data. The lowest standard deviation value is found as 0.03
for the η parameter of the Quebec field data. Very low standard deviation values are
obtained for the q and η parameters of both Camaguey and Quebec field data according
to their relatively low parameter values. On the other hand, other parameters of A, z0
ve x0, that have relatively high parameter values, are caused higher standard deviation
results than q and η parameters. The highest standard deviation value for parameter A
is found as 390.95 as a result of the inversion of the Camaguey field data with the
CSA.

Table 4.18 CSA estimation, mean and standard deviation values of Camaguey and Quebec field data
Camaguey Anomaly Quebec Anomaly
Parameters CSA Mean CSA Mean
SD SD
Estimation Value Estimation Value
A [mGal⋅m2q-η] 1430.30 1603.03 390.95 204.04 231.2317 69.05
z0 [m] 23.30 25.6470 5.43 49.03 54.1681 15.17
q 1.49 1.5164 0.12 0.92 0.9100 0.12
η 0.18 0.2254 0.05 0.23 0.2142 0.03
x0 [m] 58.72 72.6309 15.35 113.66 113.3848 1.17

4.3 Results and Discussions of Inversion Studies using CSA

An application of the CSA is presented in Section 4.1 for the inversion of the SP
anomalies based on the data sets comprising the theoretical and field examples. CSA
is applied to the theoretical and field SP anomalies including Bavarian (Germany),
Süleymanköy (Türkiye) and Malachite mine (USA) for the inversion of the model
parameters using the control parameters via the parameter tuning study using the noise-
free data set. Based on the comparisons between present study and prior studies using
different metaheuristics to invert the same data sets, it is seen that the results from CSA
show good agreements with the previous studies. The comparisons yield that there is
a good similarity among the solutions provided by the CSA, DE, PSO, and SA. A rapid
convergence in the error energies and the model parameters is also observed in CSA.
CSA requires only two algorithm-based parameters (the number of population and
probability of recognition of the egg) and this can be mentioned as another advantage
of the algorithm.

91
On the other hand, CSA is performed in Section 4.2 for the inversion of the
theoretical (without noise and noisy) and field (Camaguey and Quebec) gravity
anomalies. The control parameters of the CSA are determined by parameter tuning
studies and statistical evaluations of the estimated results. The control parameters of
the algorithm are found 40 and 0.05 for Np and ρa, respectively. The inversion studies
of theoretical (without noise and noisy) and field (Camaguey and Quebec) gravity data
are performed with 300 iterations. According to the results of the inversions of gravity
anomalies with CSA, user defined two control parameters indicate simple use for the
model parameter estimations.

It can be said that the CSA is an efficient alternative method according to its
capability of reaching global minimums without needing a good initial model with
acceptable computational time and rms values for the inversion of the SP and gravity
anomalies in geophysics. Additionally, as a result of the uncertainty analysis using the
M-H algorithm, it is seen that the model parameter estimations give reliable results for
the inversions of SP and gravity anomalies with CSA.

92
CHAPTER FIVE
BAT ALGORITHM FOR SELF-POTENTIAL AND GRAVITY ANOMALIES

Bat algorithm (BA) is a population-based metaheuristic method, which was


proposed by Yang (2010). Three main rules were defined for the BA. First, bats use
echolocation to detect distance and they know the difference between food and
obstacles. Second, bats fly randomly with a constant frequency and variable
wavelength and sound pulse in a specific location and speed. Bats can automatically
arrange the frequency (or wavelength) of the emitted pulses and pulse emission rate
according to the distance between them and their targets. Finally, the sound pulse is
assumed to change from a high value to a constant minimum value.

5.1 Echolocation Characteristics of Bats

Echolocation is used by bats as a type of radar to locate their prey, communicate


with each other, detect thin objects, and distinguish between different types of insects
at their dark environments. All animals including bats that use echolocation shrews
(dolphins, whales, and some bird species) emit signals at a certain frequency. These
signals are called ultrasonic, and they are higher than upper limits of what human ear
can hear (about 20 kHz). Bats contain a wide range of frequencies, from very high
(>200 kHz) to low (~10 kHz), which is not ultrasonic, and the human ear can easily
detect the sounds with low frequencies (Fenton, 2004).

Bats decide, categorize, and locate the objects in their environment by analyzing
the echoes as a result of the signals of the bats. Bats can decide the existing of the prey
according to the echolocation signal or their sense of smelling, hearing, seeing. On the
other hand, bats can categorize their targeting prey by the echoes generated by the
echolocation signal. Characteristics such as size, shape and texture of the targets are
contained in echoes as a complex form of code. The amplitude of the echo and the
frequency-modulated signal provide information about the movements of the prey.
Echolocation informs location information to the bats as vertical and horizontal

93
position of a target and its distance from the bat. The elapsed time of the signal from
the bat to the first time the echo returns to the bat informs the prey’s distance from the
bat (Yang, 2010).

5.2 Local Search Characteristics of Bat Algorithm

To increase the local search capability of the algorithm, it is assumed that the
individuals of the population can find better quality sources near the existing prey
sources. This system has been included the bat algorithm by Yang (2010). For this
purpose, each individual or solution that meets desired conditions, chooses another
solution depending on the quality of the fitness value from the population and searches
a new source near the old solution. The new solution can be described as follows.

𝑥𝑛𝑒𝑤 = 𝑥𝑜𝑙𝑑 + 𝜀𝐴̅𝑡 (5.1)

where, ε is a randomly generated value between -1 and 1, xold is the solution value of
the chosen individual from the population based on the quality of the fitness value by
any mechanisms such as roulette wheel or elitism, Ᾱt represents average sound pulse
of all bats at the t time.

Generated sound pulse and the rate of pulse emission of the bats with echolocation
should be updated depending on the iteration number. If the bat is getting closer to
their prey, sound pulse (A) is decreased, and rate of pulse emission is increased.

𝐴𝑡+1
𝑖 = 𝛼𝐴𝑡𝑖 (5.2)

𝑟𝑖𝑡+1 = 𝑟𝑖0 [1 − 𝑒𝑥𝑝(−𝛾𝑡)] (5.3)

where, α and γ are constant values as 0<α<1 and γ>0. Initial sound pulse (𝐴0𝑖 ) and
initial rate of pulse emission (𝑟𝑖0 ) are determined randomly and usually they can
describe as 𝐴0𝑖 [0,1] and 𝑟𝑖0 𝜖 [0,1]. According to the rules of the BA, a flowchart is
illustrated in Figure 5.1 with the number of population (Np), sound pulse (A) and pulse

94
rate (r) as the algorithm-based (control) parameters of the BA for the inversion of SP
and gravity anomalies.

Figure 5.1 Flowchart of the Bat Algorithm

As shown in the flowchart of the Bat Algorithm, random values of the r and A are
compared with the actual r and A to find the best solution for each iteration until the
iteration number reaches to the population number. If the best solution is not obtained,
new solution is achieved by increased r and reduced A parameters.

95
5.3 Inversion of SP Anomalies with Bat Algorithm

Definition of the SP, mechanisms of electrical fields for SP and research areas of
the SP methods were discussed in Section 4.1. in detail. In this section, inversion of
the SP anomalies is performed using Bat Algorithm for synthetic and field data set.
The model parameters of the SP anomaly are given as electric dipole moment (K),
shape factor (q), polarization angle (θ), the exact origin (x0) and depth (z0) of the
causative body depending on the distance in horizontal along the observation line (x)
for some simple geometrical shapes including an SGM-1, SGM-2, and SGM-3.
According to these model parameters, the mathematical model of the SP, which was
given in Eq. (4.4), is used for the inversion with Bat Algorithm.

5.3.1 Parameter Tuning of Self-Potential Data for Bat Algorithm

BA needs optimum control parameters before the inversion studies of synthetic and
field SP data sets. The control parameters of the BA are determined with a parameter
tuning study. For this reason, control parameters of the BA (A and r) are arranged from
0 to 1 with 0.1 increment for parameter tuning study. The control parameters of the
BA estimated model parameters for noise free synthetic SP anomaly, the rms values
(calculated with Eq. 4.5) of the parameters and mean elapsed time for 300 iterations in
10 independent runs are given in Table 5.1 for parameter tuning of the BA.

The lowest rms and the standard deviation value are obtained from the parameter
tuning study as 0.158 mV for mean rms, 0.098 mV for min rms and 0.229 mV for max
rms with 0.053 mV standard deviation when the control parameters A is 0.01 and the
r is 0.02. Model parameters for this control parameters, which are yielded for lowest
standard deviation, are obtained as 1006.99 [mVm2q-1] for K, 170.01 [m] for 𝜃, 30.00
for z0 [m], 1.00 for q and 249.94 for x0 [m]. The control parameters of the BA, in which
the lowest rms and standard deviations were obtained, are used in the inversion of the
theoretical data without noise.

96
Table 5.1 Parameter tuning study for BA with the theoretical data without noise for SP anomaly (The
best result of the parameter tuning study is shown in boldface type)
Model Parameters rms (mV)
A r K θ z0 x0
q Min Max Mean SD
[mVm2q-1] [°] [m] [m]
0.1 1026.33 171.09 30.72 1.01 250.06 0.201 0.412 0.311 0.105
0.2 1036.99 170.21 31.00 0.99 249.94 0.098 0.229 0.158 0.053
0.3 989.76 169.78 29.89 0.99 249.93 0.171 0.367 0.269 0.098
0.4 1100.66 169.86 31.07 0.99 249.88 0.243 0.473 0.312 0.118
0.5 977.87 169.96 29.86 0.99 249.98 0.220 0.404 0.323 0.092
0.1
0.6 965.84 169.82 29.77 0.99 249.87 0.162 0.382 0.244 0.111
0.7 1060.29 170.13 30.42 1.00 249.99 0.231 0.467 0.378 0.119
0.8 1035.59 169.89 30.11 1.00 249.89 0.225 0.523 0.382 0.149
0.9 988.87 169.99 30.00 0.99 250.04 0.161 0.379 0.271 0.109
1.0 998.71 169.98 30.04 0.99 250.09 0.115 0.314 0.246 0.101
0.1 1018.30 170.06 30.00 1.03 250.01 0.362 0.648 0.476 0.143
0.2 1030.85 170.00 30.10 1.00 249.97 0.229 0.444 0.381 0.110
0.3 995.26 170.06 29.92 0.99 250.04 0.113 0.337 0.230 0.112
0.4 1026.68 171.04 30.25 1.00 249.95 0.224 0.352 0.299 0.064
0.5 972.99 170.01 29.64 0.99 249.99 0.235 0.356 0.286 0.060
0.2
0.6 1049.10 170.07 30.37 1.00 250.01 0.126 0.253 0.219 0.064
0.7 1012.51 171.07 30.15 1.00 249.98 0.121 0.357 0.261 0.118
0.8 1144.28 170.16 29.91 1.00 250.01 0.164 0.282 0.238 0.062
0.9 965.34 170.03 29.74 0.99 250.07 0.134 0.359 0.283 0.114
1.0 1030.27 169.02 29.76 0.99 251.05 0.127 0.292 0.237 0.084
0.1 973.42 169.81 29.76 0.99 249.90 0.140 0.252 0.229 0.052
0.2 992.80 169.99 29.97 0.99 250.01 0.108 0.279 0.208 0.060
0.3 965.55 169.86 29.68 0.99 249.93 0.127 0.365 0.284 0.121
0.4 1058.34 170.17 30.45 1.00 249.99 0.236 0.348 0.293 0.056
0.5 978.70 170.02 29.91 0.99 250.05 0.222 0.341 0.257 0.061
0.3
0.6 957.33 169.97 29.88 0.99 249.99 0.218 0.441 0.333 0.111
0.7 1031.27 170.08 30.27 1.00 250.15 0.221 0.352 0.297 0.065
0.8 998.92 169.80 30.10 0.99 249.95 0.237 0.357 0.304 0.060
0.9 1038.46 170.14 30.30 1.00 250.00 0.217 0.346 0.283 0.064
1.0 1074.10 169.98 30.41 1.00 250.04 0.146 0.284 0.241 0.070
0.1 1206.13 170.32 30.05 1.01 250.15 0.139 0.266 0.237 0.066
0.2 1049.75 170.11 30.39 1.00 250.00 0.165 0.349 0.239 0.092
0.3 1037.66 170.16 30.33 1.00 250.12 0.130 0.341 0.235 0.105
0.4 970.33 170.05 29.78 0.99 250.12 0.172 0.291 0.243 0.059
0.5 1019.05 170.02 30.00 1.00 250.02 0.125 0.367 0.242 0.121
0.4
0.6 1033.23 170.04 30.23 1.00 250.05 0.223 0.360 0.314 0.069
0.7 973.65 169.40 29.90 0.99 249.03 0.236 0.349 0.285 0.056
0.8 1036.67 171.06 30.04 1.00 251.05 0.237 0.464 0.341 0.113
0.9 975.27 170.08 29.82 0.99 250.06 0.221 0.361 0.279 0.070
1.0 1059.35 171.05 31.09 1.00 249.95 0.233 0.351 0.298 0.059
0.1 980.86 169.91 29.80 0.99 250.02 0.162 0.355 0.310 0.101
0.2 1012.91 169.93 29.98 1.00 249.93 0.123 0.295 0.249 0.089
0.3 981.11 170.03 29.89 0.99 250.05 0.217 0.381 0.286 0.082
0.4 1020.66 171.24 30.19 1.00 250.12 0.230 0.448 0.345 0.109
0.5 981.71 169.86 29.82 0.99 249.96 0.219 0.349 0.286 0.065
0.5
0.6 966.79 169.90 30.03 0.99 249.94 0.223 0.437 0.313 0.107
0.7 1027.41 170.17 30.23 1.00 250.09 0.123 0.354 0.277 0.117
0.8 991.86 169.90 29.94 0.99 250.01 0.117 0.292 0.238 0.089
0.9 951.48 169.86 29.56 0.99 249.92 0.197 0.348 0.291 0.076
1.0 987.16 170.01 29.98 0.99 250.00 0.181 0.306 0.234 0.062

97
Table 5.1 Continues
0.1 1001.57 168.77 30.01 1.02 249.91 0.229 0.418 0.340 0.094
0.2 986.24 170.03 29.88 0.99 249.97 0.203 0.352 0.308 0.076
0.3 1040.39 170.06 30.11 1.00 249.92 0.217 0.383 0.244 0.089
0.4 1025.09 170.10 30.28 1.00 250.00 0.229 0.457 0.341 0.114
0.5 1207.07 169.97 30.11 1.01 250.06 0.123 0.405 0.335 0.146
0.6
0.6 983.84 170.04 29.99 0.99 250.04 0.182 0.353 0.297 0.087
0.7 989.73 169.95 30.02 0.99 249.93 0.219 0.472 0.332 0.126
0.8 974.75 169.95 29.80 0.99 249.88 0.208 0.348 0.312 0.072
0.9 1024.27 170.04 30.12 1.00 250.10 0.176 0.369 0.301 0.097
1.0 987.45 169.85 29.96 0.99 249.92 0.117 0.356 0.233 0.119
0.1 1010.89 169.92 30.14 1.00 249.89 0.207 0.342 0.264 0.067
0.2 988.82 170.10 29.88 0.99 250.17 0.134 0.357 0.278 0.113
0.3 984.84 169.90 29.94 0.99 249.88 0.222 0.321 0.247 0.051
0.4 961.33 169.90 29.92 0.99 249.94 0.136 0.413 0.279 0.138
0.5 1051.46 170.16 30.31 1.02 250.03 0.123 0.356 0.293 0.120
0.7
0.6 1102.69 170.96 29.21 1.00 250.11 0.218 0.351 0.291 0.066
0.7 1027.96 169.92 30.11 1.01 249.91 0.178 0.312 0.278 0.069
0.8 1257.51 170.25 30.40 1.00 250.11 0.210 0.386 0.299 0.088
0.9 1033.14 169.92 30.26 1.00 249.95 0.175 0.427 0.247 0.129
1.0 1004.92 170.14 30.05 1.01 250.15 0.131 0.273 0.224 0.072
0.1 1023.76 170.22 30.18 1.01 251.07 0.226 0.460 0.339 0.117
0.2 1030.55 170.08 30.22 1.00 250.02 0.218 0.452 0.302 0.118
0.3 1010.10 169.88 30.07 1.00 249.94 0.263 0.467 0.341 0.103
0.4 989.69 169.91 29.92 0.99 250.02 0.188 0.355 0.296 0.084
0.5 944.74 169.80 29.60 0.99 250.02 0.237 0.451 0.342 0.107
0.8
0.6 975.93 169.86 29.96 0.99 249.88 0.229 0.417 0.315 0.094
0.7 1031.14 169.99 30.19 1.00 249.93 0.171 0.354 0.297 0.093
0.8 1102.53 170.04 30.16 1.00 250.08 0.241 0.401 0.335 0.080
0.9 1032.55 170.11 30.26 1.00 250.06 0.187 0.396 0.293 0.104
1.0 1042.17 169.80 30.28 1.00 249.86 0.140 0.347 0.243 0.103
0.1 1035.51 170.01 30.13 1.00 250.01 0.126 0.363 0.251 0.118
0.2 1049.76 170.07 30.35 1.00 250.01 0.188 0.395 0.293 0.103
0.3 969.18 170.00 29.76 0.99 250.03 0.223 0.402 0.334 0.090
0.4 1041.01 170.14 30.12 1.00 250.04 0.231 0.419 0.312 0.094
0.5 996.22 169.93 29.99 0.99 249.87 0.125 0.360 0.271 0.118
0.9 0.6 1023.95 170.06 30.06 1.00 250.04 0.219 0.381 0.307 0.081
0.7 970.46 169.90 29.64 0.99 249.90 0.236 0.364 0.285 0.064
0.8 992.32 170.09 29.88 0.99 249.98 0.129 0.417 0.335 0.148
0.9 1024.15 170.09 30.18 1.00 250.05 0.234 0.397 0.289 0.082
1.0 1018.60 170.18 30.13 1.00 250.01 0.123 0.305 0.236 0.091
0.1 988.45 169.93 30.02 0.99 249.95 0.141 0.359 0.279 0.110
0.2 989.55 170.12 29.90 0.99 250.12 0.123 0.448 0.337 0.165
0.3 1028.58 170.10 30.32 1.00 250.04 0.117 0.355 0.282 0.121
0.4 1017.24 170.03 30.06 1.00 250.01 0.212 0.403 0.308 0.095
0.5 998.29 169.96 30.14 0.99 249.96 0.233 0.447 0.341 0.107
1.0
0.6 1046.86 169.98 30.26 1.00 249.99 0.186 0.345 0.239 0.081
0.7 936.32 169.81 29.52 0.99 249.95 0.133 0.356 0.271 0.112
0.8 997.61 169.90 29.89 0.99 249.93 0.119 0.353 0.263 0.118
0.9 1201.34 170.02 30.06 0.99 249.95 0.166 0.357 0.238 0.096
1.0 993.41 170.20 29.96 0.99 250.16 0.207 0.377 0.303 0.085

98
5.3.2 Numerical Tests with Theoretical Data for Self-Potential

Numerical tests are carried out using Bat Algorithm for both the noise-contaminated
and without noise SP theoretical data to invert the synthetic anomalies. The synthetic
noise-contaminated data set was obtained by adding pseudo-random numbers to the
without noise data set with ±2 mV standard deviation and zero-mean. True values,
parameter bounds, and estimated parameters of without noise and noise-contaminated
synthetic data are given in Table 5.2 using Bat Algorithm for the inversion study.

Table 5.2 Solutions based on the theoretical data sets of SP anomaly by Bat Algorithm
True Parameter Bounds BA Estimations
Parameters
Value Minimum Maximum Noise-free Data Noisy Data
K [mVm2q-1] 1000 -1500 1500 1036.99 1170.85
𝜃 [°] 170 0 200 170.21 171.04
z0 [m] 30 0.1 50 31.00 31.28
q 1 0.1 2.0 0.99 1.01
x0 [m] 250 50 300 249.94 250.82
rms [mV] - - - 0.098 2.040

Estimation results of the without noise and noise-contaminated data sets are
obtained between minimum and maximum bounds of the true model parameters. Low
rms results are yielded by the inversions of synthetic data with BA as 0.008 mV and
2.040 mV for without noise and noise-contaminated data respectively. According to
the noise level of ±2 mV, the rms of noise-contaminated data can be observed higher
than the rms results of the data without noise. It is understood that the BA is successful
for the model parameter estimations of synthetic SP data based on the rms results.

Inversion results of synthetic SP anomalies with BA are illustrated in Figure 5.2


including observed and calculated anomalies with the values of the results (Figure
5.2a) and the change of the error energy (Figure 5.2b) for noise-free data. The inversion
study is implemented on the synthetic SP anomalies for 300 iterations to estimate the
model parameters.

99
Figure 5.2 Inversion result of the SP noise-free data for 300 iteration (a) and error energy according to
the iteration (b) with Bat Algorithm

For each model parameter of the SP anomaly without noise, the convergence of the
achieved model parameter estimations during 300 iterations are shown in the Figure
5.3 for K (Figure 5.3a), θ (Figure 5.3b), z0 (Figure 5.3c), q (Figure 5.3d) and x0 (Figure
5.3e). Good and fast convergence characteristics are obtained especially in the
parameters of θ, z0 and x0 for synthetic data without noise using the BA.

Figure 5.3 Convergence of the model parameters for SP noise-free theoretical data according to the
iteration number (a-e)

On the other hand, as inversion results for 300 iterations to estimate the model
parameters of noise-contaminated synthetic data using BA, observed and calculated
anomalies with the results are given in Figure 5.4a. Also, the change of the error energy
for noise-contaminated SP data is shown in Figure 5.4b.

100
.

.Figure 5.4 Inversion result of the SP noisy data for 300 iteration (a) and error energy according to the
iteration (b) with Bat Algorithm

The convergences of the model parameters for each model parameter of the noise-
contaminated SP anomaly are given in the Figure 5.5 during 300 iterations for K
(Figure 5.5a), θ (Figure 5.5b), z0 (Figure 5.5c), q (Figure 5.5d) and x0 (Figure 5.5e).
For noise-contaminated synthetic data, the parameters of θ and x0 are converged fast
to their estimated parameter. However, the parameters of K and z0 in noise-
contaminated synthetic SP anomaly yielded a bit far from the true values using the BA.

Figure 5.5 Convergence of the model parameters for SP noisy theoretical data according to the iteration
number (a-e)

Error energy maps are given in the Fig. 5.6 for the noise free SP anomaly obtained
by BA. The scale of the contours is located in the right and true values of the anomaly
and estimated model parameters are also shown in the figure with black and red dots
respectively.

101
Figure 5.6 Error energy maps of SP synthetic noise-free data for model parameters using BA

102
The results of an uncertainty appraisal analysis by M-H algorithm for each model
parameter are given in Table 5.3. As a result of the uncertainty appraisal analysis,
estimated model parameters, mean values and standard deviations are listed for both
without noise and noise-contaminated synthetic SP anomalies using Bat Algorithm.

Table 5.3 Bat Algorithm estimations for SP model parameters, mean and standard deviation values of
noise-free and noisy theoretical data
Noise-free Data Noisy Data
Parameters BA Mean BA Mean
SD SD
Estimation Value Estimation Value
K [mVm2q-1] 998.52 829.66 166.18 1170.85 968.05 171.54
𝜃 [°] 169.97 169.46 5.78 171.04 170.29 5.60
z0 [m] 29.96 29.23 3.54 31.28 30.86 3.39
q 0.99 0.98 0.02 1.01 0.99 0.02
x0 [m] 250.01 249.73 4.18 250.82 250.09 4.30

Relative frequency histograms are shown in Figure 5.7 for synthetic data for each
model parameter of without noise synthetic (Figure 5.7a-e) and noise-contaminated
synthetic (Figure 5.7f-j) SP anomalies. Histograms of the relative frequency
distribution are obtained from the uncertainty analysis of M-H sampling with 90001
samples for each model parameter.

Figure 5.7 Histograms of relative frequency of SP noise-free and noisy synthetic data by M-H algorithm

103
After the inversion of the without noise and noise-contaminated synthetic data,
without noise multiple synthetic SP data are used for the inversion by BA. The multiple
synthetic SP anomaly includes two separate anomalies called Source-1 and Source-2
same as the multiple SP anomaly used in Section 4.1.4 by CSA. The results of error
energy and the details of the SP anomaly as observed and calculated values are shown
in Fig. 5.8 for 600 iteration using BA.

Figure 5.8 Error energy (a) and inversion result of multiple synthetic data for 600 iterations

True values of the multiple SP synthetic anomaly for both for Source-1 and Source-
2, parameter bounds of Source-1 and Source-2 and model parameter estimations by
BA are listed in Table 5.4. According to the inversion results of the multiple SP
synthetic anomaly, estimated model parameters are found with 3.62 mV rms value.
Additionally, changing of the model parameters by iteration are shown in the Fig. 5.9
for both for Source-1 and Source-2 of the multiple noise-free SP synthetic anomaly.

Table 5.4 True values, parameter bounds and BA estimations for the multiple synthetic data.
Bounds for Bounds for
True Values BA Estimations
Parameters Source-1 Source-2
Source-1 Source-2 Min Max Min Max Source-1 Source-2
K [mVm2q-1] -1500 115 -2000 1000 50 500 -1322.61 120.91
𝜃 [°] 165 235 0 180 90 360 165.62 236.81
z0 [m] 5 10 0 15 5 25 4.66 9.21
q 1 0.5 0 2 0 2 0.98 0.49
x0 [m] 75 275 0 200 100 500 73.22 275.73
rms [mV] - - - - - - 3.62

104
105
Figure 5.9 Change of the model parameters (a-e) Source-1 and (f-j) Source-2 for multiple synthetic data sets according to iteration number.
5.3.3 Tests with Field Data for Self-Potential

Bavarian woods in Germany, Süleymanköy in Türkiye, and Malachite mine in


Colorado (USA) were used as SP field anomaly for the inversion with field data to
investigate the applicability and the performance of the BA. These field data sets were
also used in previous Section for the inversion of field data sets using CSA.

5.3.3.1. Bavarian Woods Anomaly (Germany)

The search spaces were set as the same with the previous Section, which are K ∈
[5000, 40000], θ ∈ [-80, -40], z0 ∈ [30, 60], q ∈ [0.1, 2.0], and x0 ∈ [240, 280].
Inversion results of the Bavarian field anomaly with BA are shown in Figure 5.10a for
observed and calculated anomalies with the values of the results and Figure 5.10b for
the change of the error energy during the iteration number.

Figure 5.10 Inversion result of the Bavarian field anomaly for 300 iteration (a) and error energy
according to the iteration number (b) with BA

For Bavarian field anomaly, the convergences of each model parameter are given
in the Figure 5.11 during 300 iterations for K (Figure 5.11a), θ (Figure 5.11b), z0
(Figure 5.11c), q (Figure 5.11d) and x0 (Figure 5.11e). According to the changes in the
model parameters by the iteration, all model parameters are converged before the 250th
iteration in Bavarian field anomaly using the BA.

106
Figure 5.11 Convergence of the model parameters for Bavarian field anomaly according to the iteration
number (a-e)

The model parameter estimation results of Bavarian field data using BA are
compared with the previous studies in the literature for various metaheuristic
algorithms and the comparison of the results in the literature are shown in Table 5.5.

Table 5.5 The solution by BA in comparison with the prior studies (The Bavarian woods anomaly)
Göktürkler and Balkaya (2012) Di Maio Present
Parameters et al. (2016) Study
GA PSO SA Genetic-Price BA
2q-1
K [mVm ] 21272.91 33343.87 26257.42 39951.67 35835.98
𝜃 [°] -51.29 -48.60 -49.98 -47.30 -48.23
z0 [m] 45.03 47.59 45.99 49.50 48.08
q 0.97 1.02 0.99 1.0 1.02
x0 [m] 268.79 269.88 269.17 271.10 270.01
rms [mV] 11.40 10.95 11.09 - 10.940

According to the comparison of Bavarian field anomaly, the BA yields better results
than the other methods according to its lower rms results. It is shown that the rms and
estimation result of the model parameter from BA are similar with PSO which is one
of the most popular metaheuristic methods in geophysical inversions.

5.3.3.2. Malachite Mine Anomaly (USA)

The Malachite mine in Colorado (USA) is used as the other field example for the
inversion study using BA. The search spaces are implemented as K ∈ [-1500, 100], θ
∈ [50, 120], z0 ∈ [0.1, 50], q ∈ [0.1, 2.0] and x0 ∈ [1.0, 184] which are the same bounds

107
with the inversions using CSA in previous Section. In order to express the inversion
results of the Malachite field anomaly using BA, observed and calculated anomalies
are presented in Figure 5.12a with the values of the results. On the other hand, the
change of the error energy is shown in Figure 5.12b according to the iteration number.

Figure 5.12 Inversion result of the Malachite field anomaly for 300 iteration (a) and error energy
according to the iteration number (b) with BA

For the inversion of Malachite field anomaly using BA, changes of the model
parameters with iteration number are given in the Figure 5.13 during 300 iterations for
K (Figure 5.13a), θ (Figure 5.13b), z0 (Figure 5.13c), q (Figure 5.13d) and x0 (Figure
5.13e). According to the convergence of the model parameters by the iteration, θ and
x0 reach their estimated values after the 250th iteration in Malachite field anomaly using
the BA.

Figure 5.13 Convergence of the model parameters for Malachite field anomaly according to the iteration
number (a-e)

108
For the Malachite field data using BA, the model parameter estimations are
compared with other inversion studies using various metaheuristic methods in the
literature. The comparison results are listed in Table 5.6 for presented study with BA
and other studies including VFSA, DE and PSO.

Table 5.6 The solution by BA in comparison with the prior studies (The Malachite mine anomaly)
Biswas and Present
Balkaya (2013)
Parameters Sharma (2015) Study
VFSA* DE PSO BA
K [mVm2q-1] -224.4 -436.11 -460.56 -438.15
𝜃 [°] 80.07 83.56 83.32 83.55
z0 [m] 15.2 19.21 19.51 19.23
q 0.5 0.62 0.62 0.61
x0 [m] -0.96 92.32 92.24 92.33
rms [mV] - 6.56 6.57 6.558
*
Very Fast Simulated Annealing

The comparison of Malachite field anomaly shows that the BA yields similar results
with the other metaheuristic methods according to the rms results. An exception can
be said that the model parameter estimation results from VFSA are far from the BA
and the other methods.

5.3.3.3. Süleymanköy Anomaly (Türkiye)

The last field example for the inversion study using BA is decided Süleymanköy
Anomaly (Türkiye). The search spaces are implemented as K ∈ [-60000, 100000], θ ∈
[5, 50], z0 ∈ [20, 50], q ∈ [0.5, 1.5] and x0 ∈ [50, 100] which are the same bounds with
the inversions using CSA in previous Section. The inversion results of the
Süleymanköy field anomaly using BA are given for observed and calculated anomalies
in Figure 5.14a with the values of the results and the change of the error energy in
Figure 5.14b during the 300 iterations.

109
Figure 5.14 Inversion result of the Süleymanköy field anomaly for 300 iteration (a) and error energy
according to the iteration number (b) with BA

Figure 5.15 shows the parameter changes of the inversion of Süleymanköy field
anomaly using BA. The changes of the model parameters with iteration number are
given for 300 iterations for K (Figure 5.15a), θ (Figure 5.15b), z0 (Figure 5.15c), q
(Figure 5.15d) and x0 (Figure 5.15e). The convergence of the model parameters of θ
and x0 are better than other parameters in Malachite field anomaly using the BA.

Figure 5.15 Convergence of the model parameters for Süleymanköy field anomaly according to the
iteration number (a-e)

The model parameter estimations are compared with other inversion studies using
various metaheuristic methods in the literature for the Süleymanköy field data using
BA. The comparison results are given in Table 5.7 for presented study with BA and
other studies including ASA, VFSA, GA, PSO and SA.

110
Table 5.7 The solution by BA in comparison with the prior studies (The Süleymanköy anomaly)
Tlas and
Biswas Present
Asfahani Göktürkler and Balkaya (2012)
Parameters (2017) Study
(2008)
ASA* VFSA** GA PSO SA BA
K [mVm2q-1] -904.03 10079.9 -36724.18 -40549.97 -44771.77 -41270.19
𝜃 [°] 17.76 72.2 22.85 22.44 22.21 22.40
z0 [m] 35.41 25.4 32.68 33.11 33.59 33.22
q 1.19 1.0 1.16 1.17 1.18 1.17
x0 [m] 15.47 72.2 79.24 79.13 79.09 79.13
rms [mV] - - 12.20 12.19 12.19 12.190
*
Adaptive Simulated Annealing **Very Fast Simulated Annealing

Süleymanköy field anomaly shows that the BA yields similar results with the other
metaheuristic methods in the comparison according to the rms results and model
parameters. However, the parameter estimation results of ASA and VFSA are not close
to the results from the BA.

5.3.3.4. KTB Borehole Anomaly (Germany)

KTB Borehole Anomaly, which was mentioned before in Section 4.1.5.4, is


occurred two boreholes called Source-1 and Source-2 with 4 km and 9.1 km depth
respectively. Therefore, KTB Borehole Anomaly has two negative peaks and the
results of error energy according to the 600 iteration and comparison of the observed
and calculated anomalies are presented in Figure 5.16 for vertical cylinder SP model
as Case-1. On the other hand, Table 5.8 shows parameter bounds and estimated values
by BA of KTB Borehole Anomaly for Source-1 and Source-2 in Case-1. Additionally,
the inversion results of BA are compared with prior studies for Case-1 in Table 5.9.

Table 5.8 The results by BA (KTB Borehole Anomaly) using vertical cylinder SP model (Case-1)
Bounds for Bounds for
BA Estimations
Parameters Source-1 Source-2
Min Max Min Max Source -1 Source-2
K [mVm2q-1] 100 1000 100 6000 1291.16 3110.51
𝜃 [°] -100 180 -100 180 -81.08 -95.08
z0 [m] 0 200 0 200 77.43 169.42
q 0 2.0 0 2.0 0.61 0.69
x0 [m] 500 1800 500 1800 1203.01 576.29
rms [mV] - - - - 39.2861

111
Figure 5.16 Inversion results of BA from the KTB Borehole (Germany) field SP data set using vertical
cylinder model (Case-1) for (a) the comparison of the observed and calculated anomalies and (b) change
of the error energy

Table 5.9 The solution by BA for Case-1 in comparison with the prior studies (The KTB Borehole
anomaly)
Source Di Maio Sungkono (2020) Present
Parameters et al. (2019) Study
GPA MDE* µJADE** CSA
Source-1 K [mVm2q-1] 4669 902.12 4078.24 1291.16
𝜃 [°] 96 -99.34 -98.37 -81.08
z0 [m] 145 96.85 167.56 77.43
q 0.75 0.60 0.73 0.61
x0 [m] 1305 1294.88 1308.41 1203.01
Source-2 K [mVm2q-1] 1664 395.43 2194.95 3110.51
𝜃 [°] 79 -80.09 -80.54 -95.08
z0 [m] 111 76.73 96.41 169.42
q 0.64 0.49 0.67 0.69
x0 [m] 1967 1975.5 1969.38 576.29
* Micro-differential evolution (MDE) variant including vectorized random mutation factor (MVDE)
** Adaptive Micro-differential evolution MDE (µADE)

On the other hand, KTB Borehole Anomaly is modeled as a multiple inclined sheet
geometry as Case-2 in addition to the SGM-1. The results of error energy according to
the 600 iteration and comparison of the observed and calculated anomalies are
presented in Figure 5.17 using inclined sheet SP model for Case-2. Additionally,
model parameter estimation results and the comparisons with prior studies are given
in Table 5.10 and 5.11 respectively for Case-2.

112
Figure 5.17 The results of BA from the KTB Borehole (Germany) SP data set using inclined sheet model
(Case-2). (a) change of the error energy and (b) the comparison of the observed and calculated anomalies

Table 5.10 The inversion results by BA (KTB Borehole Anomaly) using inclined sheet SP model (Case-
2)
Bounds for Bounds for
CSA Estimations
Parameters Source-1 Source-2
Min Max Min Max Source -1 Source-2
U0 [mV] -1000 -10 -1000 -10 -67.96 -83.74
𝜃 [°] 0 180 0 180 84.64 124.15
h [m] 10 1000 10 1000 581.09 193.96
a 1 2000 1 2000 606.75 250.69
x0 [m] 0 1800 0 1800 528.65 1342.11
rms [mV] - - - - 41.92

Table 5.11 The solution by BA for Case-2 in comparison with the prior studies (The KTB Borehole
anomaly)
Source Gobashy Sungkono (2020) Present
Parameters et al. (2020) Study
WOA* MVDE** µJADE*** BA
Source-1 U0 [mV] 58.63 74.98 67.08 -67.96
𝜃 [°] 40.22 79.11 80.00 84.64
h [m] 468.53 447.09 556.99 581.09
a [m] 739.25 429.13 530.23 606.75
x0 [m] 632.70 505.52 500.00 528.65
Source-2 U0 [mV] 54.53 120.06 91.67 -83.74
𝜃 [°] 153.76 103.95 117.76 124.15
h [m] 308.21 134.82 154.15 193.96
a [m] 655.92 128.86 149.84 250.69
x0 [m] 1173.56 1224.99 1271.08 1342.11
*
Whale optimization algorithm
**
Micro-differential evolution (MDE) variant including vectorized random mutation factor (MVDE)
***
Adaptive Micro-differential evolution MDE (µJADE)

113
According to the results of SP inclined sheet model from the inversion results of
KTB Borehole Anomaly using BA, subsurface structure of the multiple anomalies of
KTB Borehole is illustrated in Fig. 5.18 under the graphic of observed (red) and
calculated (blue) anomalies including related model parameters.

Figure 5.18 Subsurface structure for multiple inclined sheet model of the KTB Borehole (Germany)
according to the calculated anomalies by BA

5.4. Inversion of Gravity Anomalies with Bat Algorithm

Gravity anomaly and definition of the mathematical model of the gravity anomaly
were discussed in Section 4.2.1 in detail. In this section, inversion of the gravity
anomalies is performed using Bat Algorithm for without noise and noise-contaminated
theoretical and field data set. The model parameters of the gravity anomaly are given
as amplitude coefficient (A), shape factors (q and η), the exact origin (x0) and depth
(z0) of the causative body depending on the distance in horizontal along the observation
line (x) for some simple geometrical shapes including a SGM-1, SGM-2, and SGM-3
(also shown in Fig. 4.13). According to these model parameters, the mathematical

114
model of the gravity, which was given in Eq. (4.6), is used for the inversion studies
with Bat Algorithm.

5.4.1 Parameter Tuning of Gravity Data for Bat Algorithm

Before the inversion studies of synthetic and field gravity data sets, determination
of the control parameters of the BA must be performed with a parameter tuning study.
Therefore, the variations of the control parameters (A and r) of the Bat algorithm are
set in the range of their lower value (0) and the upper value (1) with 0.1 increment for
parameter tuning study. Table 5.12 shows the control parameters, model parameters,
the rms values of the parameters and mean elapsed time in 10 independent runs with
BA. Each run is performed for 300 iterations. The control parameters of the BA are
determined as a result of these runs depending on rms (Eq. 4.5) and elapsed time.

According to the results of the parameter tuning study for the theoretical gravity
data without noise, the lowest rms and the standard deviation value are obtained as
0.012 mGal for mean rms, 0.007 mGal for min rms, 0.017 mGal for max rms with
0.004 mGal standard deviation for the control parameters of the BA. The values of the
control parameters are found as 0.5 for A and 0.2 for r. On the other hand, estimated
model parameters according to the control parameters are obtained as 534.17
[mGal⋅m2q-η] for A, 10.01 [m] for z0, 1.50 for q, 0.97 for η and 39.99 for x0 [m].

115
Table 5.12 Parameter tuning study for BA with the theoretical data without noise for gravity anomaly
(The best result of the parameter tuning study is shown in boldface type)
Model Parameters rms (mGal)
A r A z0 x0
q η Min Max Mean SD
[mGalm2q-η] [m] [m]
0.1 595.80 10.01 1.51 0.94 39.99 0.028 0.049 0.038 0.009
0.2 624.55 9.95 1.48 0.87 40.01 0.033 0.050 0.043 0.007
0.3 422.34 10.01 1.49 1.07 40.01 0.024 0.063 0.042 0.017
0.4 643.99 9.94 1.48 0.86 39.99 0.022 0.038 0.031 0.006
0.5 800.27 10.03 1.51 0.81 40.01 0.017 0.035 0.025 0.007
0.1
0.6 575.86 10.05 1.51 0.96 40.00 0.019 0.051 0.036 0.012
0.7 479.00 10.04 1.50 1.03 40.00 0.017 0.053 0.029 0.014
0.8 953.63 10.01 1.51 0.73 39.99 0.027 0.045 0.037 0.008
0.9 610.47 10.02 1.50 0.92 39.99 0.014 0.060 0.040 0.017
1.0 748.15 9.95 1.49 0.81 39.99 0.029 0.047 0.039 0.006
0.1 556.11 9.98 1.49 0.94 40.00 0.020 0.024 0.022 0.002
0.2 780.73 10.01 1.50 0.81 39.99 0.010 0.019 0.014 0.004
0.3 719.37 9.98 1.49 0.83 40.00 0.013 0.032 0.020 0.007
0.4 738.29 10.00 1.49 0.82 39.99 0.017 0.027 0.021 0.004
0.5 617.13 9.99 1.50 0.91 39.99 0.009 0.027 0.018 0.00
0.2
0.6 679.48 10.03 1.50 0.88 40.00 0.011 0.035 0.020 0.009
0.7 392.62 10.02 1.50 1.11 40.01 0.028 0.045 0.035 0.006
0.8 448.80 9.99 1.50 1.04 39.99 0.011 0.036 0.023 0.010
0.9 633.94 10.04 1.50 0.91 40.00 0.015 0.039 0.024 0.009
1.0 695.26 9.99 1.50 0.85 40.00 0.013 0.044 0.028 0.011
0.1 623.59 10.02 1.50 0.91 40.00 0.007 0.027 0.019 0.008
0.2 749.35 9.94 1.48 0.79 40.00 0.017 0.044 0.029 0.010
0.3 609.29 9.98 1.49 0.90 39.99 0.005 0.056 0.022 0.020
0.4 423.93 10.02 1.50 1.07 40.00 0.015 0.042 0.032 0.010
0.5 632.71 9.99 1.49 0.89 39.99 0.010 0.027 0.022 0.007
0.3
0.6 566.40 9.98 1.49 0.94 40.00 0.018 0.033 0.025 0.007
0.7 587.24 10.05 1.51 0.95 40.01 0.017 0.034 0.025 0.006
0.8 612.56 10.03 1.50 0.92 39.99 0.020 0.085 0.038 0.026
0.9 636.98 9.98 1.49 0.89 39.99 0.009 0.036 0.017 0.011
1.0 414.51 10.00 1.49 1.08 39.99 0.006 0.137 0.057 0.052
0.1 463.31 10.01 1.49 1.03 39.98 0.028 0.071 0.041 0.017
0.2 716.12 10.06 1.51 0.88 40.00 0.036 0.092 0.061 0.024
0.3 741.54 10.02 1.50 0.83 39.99 0.024 0.085 0.045 0.024
0.4 439.62 10.00 1.50 1.05 39.99 0.012 0.094 0.061 0.031
0.5 547.81 10.02 1.50 0.97 39.99 0.025 0.073 0.041 0.019
0.4
0.6 661.14 9.99 1.49 0.87 40.00 0.013 0.071 0.033 0.022
0.7 780.42 10.00 1.50 0.81 40.00 0.022 0.058 0.033 0.015
0.8 632.01 9.95 1.49 0.87 39.99 0.013 0.080 0.038 0.026
0.9 584.90 9.95 1.49 0.90 39.99 0.023 0.058 0.047 0.013
1.0 600.46 10.01 1.50 0.92 40.00 0.014 0.035 0.019 0.008
0.1 747.35 10.03 1.51 0.84 40.00 0.014 0.027 0.021 0.006
0.2 534.17 10.01 1.50 0.97 39.99 0.007 0.017 0.012 0.004
0.3 790.58 9.99 1.49 0.79 40.00 0.006 0.024 0.016 0.009
0.4 939.96 9.98 1.49 0.71 40.00 0.006 0.018 0.013 0.005
0.5 505.27 9.98 1.49 0.99 40.00 0.008 0.028 0.015 0.008
0.5
0.6 657.79 10.00 1.50 0.88 40.00 0.013 0.018 0.016 0.002
0.7 675.44 10.01 1.50 0.87 39.99 0.011 0.016 0.013 0.002
0.8 540.37 10.02 1.50 0.97 40.00 0.010 0.031 0.018 0.008
0.9 557.65 9.98 1.49 0.94 39.99 0.013 0.027 0.021 0.006
1.0 831.96 9.93 1.49 0.75 39.99 0.032 0.131 0.079 0.035

116
Table 5.12 Continues
0.1 532.70 10.01 1.50 0.99 40.01 0.040 0.093 0.066 0.021
0.2 432.02 10.12 1.51 1.11 39.98 0.047 0.072 0.059 0.012
0.3 818.65 10.00 1.49 0.78 40.01 0.054 0.144 0.091 0.036
0.4 109.99 10.18 1.52 1.72 40.01 0.064 0.092 0.077 0.009
0.5 615.32 9.88 1.47 0.85 39.99 0.036 0.067 0.051 0.013
0.6
0.6 333.18 10.06 1.51 1.21 39.97 0.048 0.082 0.067 0.013
0.7 606.00 10.00 1.50 0.93 40.00 0.038 0.105 0.063 0.025
0.8 639.49 9.95 1.49 0.88 40.01 0.052 0.075 0.066 0.011
0.9 583.38 9.98 1.49 0.93 39.99 0.034 0.112 0.064 0.028
1.0 724.36 9.74 1.45 0.73 39.99 0.074 0.106 0.091 0.012
0.1 331.07 10.02 1.51 1.21 40.01 0.054 0.074 0.064 0.007
0.2 854.29 10.02 1.51 0.78 39.98 0.044 0.121 0.074 0.034
0.3 520.71 10.19 1.53 1.06 40.00 0.057 0.098 0.081 0.018
0.4 357.08 9.97 1.48 1.12 39.98 0.048 0.075 0.060 0.012
0.5 620.78 9.87 1.47 0.85 39.98 0.044 0.116 0.076 0.027
0.7
0.6 246.50 9.94 1.49 1.28 39.97 0.067 0.114 0.086 0.018
0.7 845.55 9.91 1.48 0.72 39.99 0.036 0.101 0.071 0.028
0.8 406.58 9.93 1.48 1.05 39.98 0.037 0.105 0.074 0.026
0.9 712.53 10.11 1.51 0.88 39.99 0.049 0.120 0.082 0.034
1.0 635.52 9.88 1.47 0.84 39.98 0.040 0.105 0.071 0.023
0.1 421.38 10.08 1.51 1.11 39.99 0.029 0.113 0.076 0.031
0.2 579.36 10.18 1.53 1.01 40.00 0.047 0.105 0.071 0.023
0.3 722.27 10.24 1.54 0.94 39.99 0.069 0.121 0.093 0.022
0.4 421.42 10.04 1.50 1.08 40.01 0.029 0.081 0.048 0.021
0.5 860.93 10.06 1.50 0.77 40.02 0.048 0.134 0.102 0.034
0.8
0.6 603.82 9.95 1.50 0.92 39.99 0.064 0.133 0.105 0.027
0.7 89.80 9.94 1.48 1.72 39.97 0.051 0.162 0.079 0.046
0.8 526.98 9.84 1.47 0.91 40.01 0.060 0.125 0.091 0.030
0.9 489.89 10.08 1.51 1.05 39.98 0.033 0.125 0.082 0.033
1.0 616.17 10.22 1.52 0.97 39.99 0.097 0.192 0.159 0.036
0.1 416.47 10.12 1.52 1.13 39.98 0.053 0.155 0.114 0.037
0.2 733.84 10.11 1.51 0.87 39.99 0.040 0.141 0.105 0.044
0.3 283.63 10.03 1.49 1.24 40.01 0.062 0.153 0.119 0.036
0.4 834.01 9.86 1.48 0.73 40.00 0.058 0.213 0.144 0.074
0.9 0.5 771.45 10.17 1.52 0.88 40.01 0.056 0.114 0.082 0.021
0.6 662.17 10.11 1.51 0.91 39.99 0.071 0.195 0.112 0.048
0.7 921.88 9.96 1.49 0.72 39.98 0.028 0.240 0.130 0.079
0.8 100.96 9.93 31.48 1.66 39.99 0.020 0.142 0.088 0.044
0.9 860.16 10.05 1.50 0.78 39.97 0.056 0.241 0.143 0.070
1.0 593.72 9.99 1.49 0.92 40.00 0.004 0.040 0.014 0.014
0.1 528.79 10.00 1.50 0.97 40.00 0.012 0.034 0.019 0.016
0.2 637.62 9.99 1.49 0.89 39.99 0.004 0.017 0.010 0.009
0.3 764.84 10.00 1.49 0.81 40.00 0.014 0.028 0.018 0.011
0.4 708.71 9.99 1.49 0.84 40.00 0.005 0.030 0.015 0.011
0.5 660.53 10.01 1.50 0.88 40.00 0.005 0.017 0.013 0.005
1.0
0.6 439.08 9.98 1.49 1.05 40.00 0.004 0.041 0.019 0.014
0.7 551.27 9.99 1.49 0.95 39.99 0.005 0.028 0.014 0.010
0.8 919.83 10.00 1.50 0.73 39.99 0.001 0.033 0.016 0.011
0.9 493.31 10.00 1.50 1.00 39.99 0.002 0.038 0.016 0.013
1.0 271.10 9.96 1.49 1.25 40.00 0.010 0.019 0.014 0.004

117
5.4.2 Numerical Tests with Theoretical Data for Gravity

For the inversion of gravity, numerical tests are performed using Bat Algorithm for
both the noise-contaminated and without noise synthetic data to estimate model
parameters. Inversion results of the true values, parameter bounds and estimated
parameters for these data are given in Table 5.13 using Bat Algorithm.

Table 5.13 Solutions based on the theoretical data sets of gravity anomaly by Bat Algorithm
True Parameter Bounds BA Estimations
Parameters
Values Minimum Maximum Noise-free Data Noisy Data
A [mGal⋅m2q-η] 500 1 1000 492.41 615.88
z0 [m] 10 1 40 9.99 11.00
q 1.5 0 2 1.50 1.64
η 1 0 2 1.01 1.27
x0 [m] 40 0 80 40.00 40.06
rms [mGal] - - - 0.001 0.256

The results of the parameter estimation using BA show that the without noise and
noise-contaminated data sets are in the range of minimum and maximum bounds of
the true model parameters. For noise-free data, rms results are obtained as 0.001 mGal
by the inversions of synthetic data with BA. On the other hand, 0.256 mGal rms are
obtained for noise-contaminated synthetic data with BA. Inversion results of the most
of the estimated model parameters (z0, q, η and x0) are close to their true values for
both without noise and noise-contaminated with pseudo random number. However,
the model parameter A is not close to its true parameter according to the inversion of
without noise and noise-contaminated synthetic data with BA. Despite the differences
between estimated value and true value of the A, BA provide low rms results for
synthetic gravity data.

Observed and calculated anomalies with the values of the results (Figure 5.19a) and
the change of the error energy (Figure 5.19b) are given for the inversion results of
noise-free synthetic anomalies with BA. In the inversion study, 300 iteration is used
for the gravity anomalies to estimate the model parameters.

118
Figure 5.19 Inversion result of the gravity noise-free data for 300 iteration (a) and error energy according
to the iteration (b) with Bat Algorithm

Convergence of the model parameters for the without noise gravity anomaly are
shown in the Figure 5.20 for A (Figure 5.20a), z0 (Figure 5.20b), q (Figure 5.20c), η
(Figure 5.20d) and x0 (Figure 5.20e) during 300 iterations. Good and fast convergence
are obtained for x0 using the BA. The other parameters are converged to estimated
values just before the 250th iteration.

Figure 5.20 Convergence of the model parameters for gravity noise-free theoretical data according to
the iteration number (a-e)

For the inversion of noise-contaminated synthetic data, observed and calculated


anomalies with the results are shown in the Figure 5.21a according to the distance of
the gravity anomaly. On the other hand, the change of the error energy is shown in
Figure 5.21b according to the iteration number for the inversion results of noise-
contaminated synthetic anomalies to estimate the model parameters with BA.

119
Figure 5.21 Inversion result of the gravity noisy data for 300 iteration (a) and error energy according to
the iteration (b) with Bat Algorithm

Changes of the values of estimated model parameters during the 300 iterations are
shown in the Figure 5.22 for A (Figure 5.22a), z0 (Figure 5.22b), q (Figure 5.22c), η
(Figure 5.22d) and x0 (Figure 5.22e) for noise-contaminated gravity data. According
to the convergence of the model parameters, good and fast convergence are obtained
only for x0 of noise-contaminated synthetic data using the BA. It can be seen that bad
convergence is yielded for the model parameters of A compared by z0, q and x0 even if
its estimated value is close to the true value. On the other hand, both bad convergence
and relatively bad estimated value are obtained for the η parameter with BA.

Figure 5.22 Convergence of the model parameters for gravity noisy theoretical data according to the
iteration number (a-e)

120
Figure 5.23 Error energy maps of gravity synthetic noise-free data for model parameters using BA

121
Error energy maps with black and red dots for the true values of the anomaly and
estimated model parameters respectively are presented in the Fig. 5.23 for the noise
free gravity anomaly obtained by BA. Additionally, the scale of the contours is located
in the right side at the figure.

The results of an uncertainty appraisal analysis by M-H algorithm for each model
parameter are given in Table 5.14 for noise free and noisy data obtained by BA. As a
result of the uncertainty appraisal analysis, estimated model parameters, mean values
and standard deviations are listed for both without noise and noise-contaminated
synthetic gravity anomalies.

Table 5.14 Bat Algorithm estimations for gravity model parameters, mean and standard deviation values
of noise-free and noisy theoretical data
Noise-free Data Noisy Data
Parameters BA Mean BA Mean
SD SD
Estimation Value Estimation Value
A [mGal⋅m2q-η] 492.41 551.26 125.48 615.88 657.83 87.74
z0 [m] 9.99 11.24 2.26 11.00 10.50 1.85
q 1.50 1.71 1.71 1.64 1.41 0.15
η 1.01 1.24 0.24 1.27 1.51 0.33
x0 [m] 40.00 39.89 4.51 40.06 38.14 4.37

Relative frequency histograms, which are obtained from the uncertainty analysis of
M-H sampling, are shown in Figure 5.24 for synthetic data for each model parameter
of without noise synthetic (Figure 5.24a-e) and noise-contaminated synthetic (Figure
5.24f-j) gravity anomalies. M-H uncertainty analysis is performed using 90001
samples for each model parameter. The distributions of the model parameters show
acceptable distribution with BA for both without noise and noise-contaminated
synthetic data.

122
FigureFigure
5.24 Histograms of relative frequency of gravity noise-free and noisy synthetic data by M-H
algorithm

5.4.3 Tests with Field Data for Gravity

5.4.3.1 Camaguey Anomaly (Cuba)

Camaguey field anomaly (Cuba) is used for the estimation of the model parameters
for an example of field data using BA. Search space of the parameters are implemented
in the tests as A ∈ [1, 10000], z0 ∈ [1, 100], q ∈ [0.5, 1.5], η ∈ [0, 1] and x0 ∈ [0, 120]
which are the same bounds with the inversions using CSA in Section 4.2.4.1. The
inversion results are presented as observed and calculated anomalies with the values
of the results (Figure 5.25a) and the change of the error energy (Figure 5.25b) for the
Camaguey field anomaly using BA during the 300 iterations.

Figure 5.25 Inversion result of the Camaguey field anomaly for 300 iteration (a) and error energy
according to the iteration number (b) with BA

123
The convergences of each model parameter are given in the Figure 5.26 for
Camaguey field anomaly during 300 iterations for the model parameters of A (Figure
5.26a), z0 (Figure 5.26b), q (Figure 5.26c), η (Figure 5.26d) and x0 (Figure 5.26e). The
graphics show that the parameters of z0, q and x0 converged fast for the estimated value
but other parameters (A and η) start to converge after the 250th iteration in Camaguey
field anomaly with the BA.

Figure 5.26 Convergence of the model parameters for Camaguey field anomaly according to the
iteration number (a-e)

Relative frequency distributions of the model parameters of Camaguey field data


sets are shown in Figure 5.27a-e. The uncertainty analysis is performed using 90001
M-H sampling for each model parameter to determine the distribution of the model
parameters which is obtained by BA.

Figure 5.27 Histograms of relative frequency of Camaguey field data by M-H algorithm

The estimated model parameters in the presented study with BA are compared with
the values taken from the other inversion studies using various metaheuristic methods

124
in the literature for the Camaguey field data. The comparison results are given in Table
5.15 for Camaguey field data sets with BA.

Table 5.15 The solution by BA in comparison with the prior studies (The Camaguey anomaly)
Biswas Ekinci et Ekinci et Present
Model (2015) al. (2016) al. (2020) Study
Parameters
VFSA DE DSA BA
A (mGal⋅m2q-η) 3.5 288.25 175.02 1314.70
z0 (m) 16.2 23.23 23.23 23.19
q 1 1.5 1.5 1.49
η - 0.71 0.86 0.21
x0 (m) -1.8 58.73 58.73 58.71
rms (mGal) 8.1∙10-3 43∙10-4 4.3e-3 0.004

5.4.3.2 Quebec Anomaly (Canada)

Inversion of Quebec field anomaly (Canada) is performed as a field example using


BA to estimate the model parameters. The search spaces are implemented in the tests
as A ∈ [1, 10000], z0 ∈ [1, 100], q ∈ [0.5, 1.5], η ∈ [0, 1] and x0 ∈ [0, 240] which are
the same bounds with the inversions using field examples with CSA in Section 4.2.4.2.
The inversion results are presented as observed and calculated anomalies with the
values of the results (Figure 5.28a) and the change of the error energy (Figure 5.28b)
for the Quebec field anomaly using BA during the 300 iterations.

Figure 5.28 Inversion result of the Quebec field anomaly for 300 iteration (a) and error energy according
to the iteration number (b) with BA

The convergences of each model parameter are given in the Figure 5.29 for Quebec
field anomaly during 300 iterations for the model parameters of A (Figure 5.29a), z0
(Figure 5.29b), q (Figure 5.29c), η (Figure 5.29d) and x0 (Figure 5.29e). The graphics

125
show that the parameter of x0 converged fast for the estimated value but other
parameters (z0, q, A and η) start to converge after the 250th iteration in Quebec field
anomaly with the BA.

Figure 5.29 Convergence of the model parameters for Quebec field anomaly according to the iteration
number (a-e)

Figure 5.30a-e shows the distributions of relative frequency for the estimated model
parameters of Quebec field data sets. To determine the distribution of the results of
model parameters obtained by BA, 90001 M-H sampling are used for the uncertainty
analysis for each model parameter.

Figure 5.30 Histograms of relative frequency of Quebec field data by M-H algorithm

The estimated model parameters in the presented study with BA are compared with
the values taken from the other inversion study using various metaheuristic methods
in the literature for the Quebec field data. The comparison results are given in Table
5.16 for Quebec field data sets with BA.

126
Table 5.16 The solution by BA in comparison with the prior studies (The Quebec anomaly)
Ekinci vd. (2016) Present Study
Model Parameters
DE BA
A (mGal⋅m2q-η) 299.11 50.33
z0 (m) 35.39 35.08
q 0.74 0.72
η 0.04 0.49
x0 (m) 113.93 113.75
rms (mGal) 0.029 0.030

5.5 Results and Discussions of Inversion Studies using BA

Details of the Bat Algorithm as echolocation and local search characteristics are
presented in Section 5.1 and 5.2 for the integration of BA to inversion studies. First,
the inversion of the SP anomalies of theoretical and field data were performed by BA
in Section 5.3. Optimum control parameters of BA are obtained by the parameter
tuning study using the noise-free data set. As field SP anomalies, Bavarian (Germany),
Süleymanköy (Türkiye) and Malachite mine (USA) data were used for the inversion
of the model parameters. Convergences of error energies and the model parameters are
observed fast with BA especially in noise-contaminated data and field examples in SP
anomaly. The comparisons between present study and prior studies using different
metaheuristics for the same SP field anomalies show that BA yields the lowest rms
values for Bavarian and Malachite anomalies. For Süleymanköy SP filed anomaly, the
comparison yields a good similarity among the solutions provided by the GA, PSO,
and SA.

On the other hand, BA is performed in Section 5.4 for the inversion of the without
noise and noise-contaminated theoretical and field gravity anomalies from Camaguey
and Quebec. The control parameters of the BA are found by parameter tuning study
performed from noise free theoretical gravity anomaly. The inversion studies of
theoretical (noise-free and noisy) and field (Camaguey and Quebec) gravity data are
performed with 300 iterations in which iteration number, characteristic of noisy data
set and field examples are the same as pervious section (CSA application) to compare
the results and define advantages or disadvantages of the algorithms. According to the
results of the inversions of gravity anomalies with BA, convergence of the parameters
is better in noise-contaminated data than noise free data. Additionally, in Camaguey

127
field example, better convergence is observed than Quebec filed example. However,
rms value in Camaguey data is lower than the other studies’ results.

128
CHAPTER SIX
A NEW APPROACH: PERTURBATION-BASED EVOLUTIONARY
ALGORITHM

Metaheuristics are the approaches aiming to find the solution of problem directly
in a pre-defined parameter space through a search method based on a stochastic
(random) process. In this context, a metaheuristic method can be defined as a strategy
guided searching process. In general, the most significant property of the nature-
inspired metaheuristics is that they do not require a well-constructed initial model.
Thus, metaheuristics as derivative-free algorithms have ability to escape from local
optima, and they usually provide global or near-global optimum solutions in a
reasonable amount of time.

6.1 Characteristics of the Perturbation-Based Evolution Algorithm

Perturbation-Based Evolutionary Algorithm (PEA) is proposed as a metaheuristic


method for the solution of the optimization problems such as inversion studies of
various anomalies in geophysics. The algorithm is mainly based on the evolution of
randomly generated initial population (i.e., candidate solutions) covering the entire
search space via mutation, crossover, and selection (i.e., genetic operators) similar to
those in DE algorithm at each generation. However, the mutation operator, which plays
the most crucial role for the success of DE, is mainly achieved by adding a scaled
difference between two individuals, which are randomly chosen from the population
to the third one. In the proposed PEA, it is performed by adding small amounts of
randomly generated perturbations to the model parameters at each generation.
According to the operating mechanisms of the PEA, the simplified operational
methodology of the proposed algorithm is shown in Fig. 6.1.

129
Figure 6.1 Simplified flowchart of the PEA

Control parameters in the PEA algorithm should be determined by a user


consequently. These parameters can significantly affect the success of the algorithm.
Control parameters and explanations of the PEA are listed in Table 6.1.

Table 6.1 Control parameters and explanations of the PEA


Parameter Explanation

Number of Generation (Ng) The maximum number of generations


Number of Population (Np) Total number of individuals, each of which represents a
solution vector of a parameter
Crossover Probability (Cr) The probability of subjecting an individual, selected
within the population, to a crossover operation thus
acquiring new generations
(Cr  [0,1])
Number of Parameter Perturbation The lower and upper limits of the perturbation to be
(permin and permax) applied for each candidate solution

130
In the PEA algorithm, an initial population is created in the following way, as in
many other metaheuristic approaches. The equation of the solution vector of the PEA
is shown in below:

𝑝(𝑖,𝑗) = 𝑝𝑎𝑟𝑟𝑎𝑛𝑔𝑒𝑚𝑖𝑛 + 𝑟 ∙ (𝑝𝑎𝑟𝑟𝑎𝑛𝑔𝑒𝑚𝑎𝑥 − 𝑝𝑎𝑟𝑟𝑎𝑛𝑔𝑒𝑚𝑖𝑛 ) (6.1)

The individuals, created in the p parameter space limits, and each of them represents
a solution vector (NxD), i and j indicate index of individuals and model parameters, r
is random corresponds to a uniform distribution of numbers of [0,1], parrangemin and
parrangemax are the upper and lower values for search space of the model parameters.

The calculation of fitness value for each candidate solution has to perform for the
determination of error energy after the initial solution. The equation of error energy is
given in Eq. 6.2 for this purpose with the PEA.

1 𝑜𝑏𝑠 2
E = ∑𝑁
𝑖=1(𝑉𝑖 − 𝑉𝑖𝑐𝑎𝑙 ) (6.2)
𝑁

where, N is the total number of data sets, and Vi is the value of the anomaly observed
or calculated (Nx1).

The PEA aims to provide a better solution at each generation by applying the
genetic operators including mutation, crossover, and selection to individuals of initial
population generated randomly. In the algorithm mutation is achieved by an approach
based on perturbation of the model parameters at each generation. Besides, mutation
and crossover operators are applied for each candidate solution unlike DE algorithm.
Mutation operator in the suggested algorithm is carried out as follows.

𝑚𝑝(𝑖,𝑗) = 𝑝(𝑖,𝑗) ∙ [1 + 𝑠𝑖𝑔𝑛(1 − 2𝑟) ∙ (𝑝𝑒𝑟𝑚𝑖𝑛 + 𝑟(𝑝𝑒𝑟𝑚𝑎𝑥 − 𝑝𝑒𝑟𝑚𝑖𝑛 ))] (6.3)

where, permin and permax are the lower and upper limits of the perturbation, respectively
which will be applied for each candidate solution; i and j indicate index of individuals
and parameters, respectively; p is jth parameters of ith individual, and r represents a

131
uniformly distributed random variable in the range of [0,1]. The PEA algorithm only
needs two user-defined control parameters consisting of permin and permax, and the
selection is carried out by an elitist approach.

On the other hand, the use of crossover and selection operators in the PEA are
shown in two following equations:

𝑚𝑝 𝑖𝑓 𝑟 ≤ 𝐶𝑟 𝑜𝑟 𝑗 = 𝑗𝑟𝑎𝑛𝑑
𝑥𝑝(𝑖,𝑗) = { (𝑖,𝑗) 𝑝(𝑖,𝑗) 𝑒𝑙𝑠𝑒𝑖𝑓 (6.4)

𝑥𝑝(𝑖,𝑗) 𝑖𝑓 𝑓(𝑥𝑝) ≤ 𝑓(𝑝)


𝑛𝑒𝑤
𝑝(𝑖,𝑗) ={ 𝑜𝑙𝑑 (6.5)
𝑝(𝑖,𝑗) 𝑒𝑙𝑠𝑒𝑖𝑓

where, mp is a perturbed individual, xp is an individual obtained in the consequence


of crossover and f is the objective function.

Classical test functions illustrated in Table 6.2 in the optimization theory are tested
for the success of the proposed metaheuristic algorithm. Main test functions of sphere,
Rosenbrock’s function, Rastrigin’s function, Beale and Himmelblau function are
performed by PEA for the estimation of true parameters of the functions. Equations,
search space bounds and graphics of each function are shown in the Table 6.2.

PEA is tested with given test functions using the control parameter as 10 for max
number of generation (Ng), 100 for population size (Np), 0.02 for crossover probability
(Cr), and 10 for permin and 100 for permax respectively.

132
Table 6.2 Details of main test functions
Search
Function Equation Visual
Space

𝐷
-5≤x1≤5
Sphere 𝑓(𝑥) = ∑ 𝑥𝑖2
-5≤x2≤5
𝑖=1

𝐷−1
2 -10≤x1≤10
Rosenbrock’s 𝑓(𝑥) = ∑ [100(𝑥𝑖2 − 𝑥𝑖+1 ) + (𝑥𝑖 − 1)2 ]
-10≤x2≤10
𝑖=1

𝐷
-5≤x1≤5
Rastrigin’s 𝑓(𝑥) = 10𝐷 + ∑[𝑥𝑖2 − 10𝑐𝑜𝑠(2𝜋𝑥𝑖 )]
-5≤x2≤5
𝑖=1

𝑓(𝑥) = (1.5 − 𝑥1 + 𝑥1 𝑥2 )2
-5≤x1≤5
Beale +(2.25 − 𝑥1 + 𝑥1 𝑥22 )2
-5≤x2≤5
+ (2.625 − 𝑥1 + 𝑥1 𝑥23 )2

𝑓(𝑥) = (𝑥12 + 𝑥2 − 11)2 -5≤x1≤5


Himmelblau
+(𝑥1 + 𝑥22 − 7)2 -5≤x2≤5

6.2 Perturbation-Based Evolutionary Algorithm with Sphere Test Function

The Sphere test function (Dixon and Szego, 1978) is continuous and unimodal
mathematical function. It has 2 local minima for two-dimensional equation and one
global minima of the sphere. The global minimum point is located in (0,0) for sphere
test function.

133
Test results of the sphere test function by PEA are given in Fig. 6.2 from initial
generation (Ng=1) to 10 generation number. True values (global minimum point) of
the sphere test function can be easily estimated as (0,0) for sphere with PEA.

Figure 6.2 Solutions of the sphere test function by PEA with 10 generation

6.3 Perturbation-Based Evolutionary Algorithm with Rosenbrock’s Test


Function

The mathematical function of the Rosenbrock, which can be described as the Valley
or Banana function, is a common test function for gradient-based optimization
algorithms (Dixon and Szego, 1978). Because the global minimum of the unimodal
function occurs in a narrow and parabolic valley. The equation, search space bounds
and graphical representation are shown in the Table 6.2 in its two-dimensional form.
Global minimum point is located in (1,1) for Rosenbrock function. In optimization
problems, despite the parabolic valley occurred by the optimum points can be found
easily, convergence to minimum points is difficult in various optimization methods.

134
Test results of the Rosenbrock’s test function by PEA are shown in Fig. 6.3 for
initial generation (Ng=1), Ng=3, Ng=5 and Ng=10 generation number. As seen in the
test results, PEA can find the valley easily but convergence to the minimum is not
exact as much as sphere test function. Finally, global minimum point is found in
(1.018, 1.035) location by PEA for Rosenbrock’s test function.

Figure 6.3 Solutions of the Rosenbrock’s test function by PEA with 10 generation

6.4 Perturbation-Based Evolutionary Algorithm with Rastrigin’s Test Function

The Rastrigin function is defined as highly multimodal, and it has several local
minimum points. However, the locations of the minimum points of the function are
regularly distributed. Equation of the test function, search space bounds and graphical
illustration of the test function of Rastrigin are shown in the Table 6.2 in its two-
dimensional form. This test function is evaluated as a hypercube, which xi ∈ [-5.12,
5.12], for i = 1 and 2, in this study. Global minimum point is located in (0,0) for
Rastrigin’s test function.

135
Test results of the Rastrigin’s test function are presented in Fig. 6.4 for initial
generation (Ng=1), Ng=3, Ng=7 and Ng=10 generation number by PEA. Test results
show that PEA can easily find the global minimum point in (0,0) location and converge
the minimum point at 7. generation for Rastrigin’s test function.

Figure 6.4 Solutions of the Rastigin’s test function by PEA with 10 generation

6.5 Perturbation-Based Evolutionary Algorithm with Himmelblau’s Test


Function

The Himmelblau’s function is a multi-model function for mathematical


optimizations which has four local minimum and one local maximum. The search
space bounds and graphical details of peak points of the Himmelblau’s test function
are shown in the Table 6.2 in its two-dimensional form. The test function have identical
four minimum local points as 0 at the points of (3,2), (-2.8051,3.1313), (-3.7793,-
3.2831) and (3.5844,-1.8481). In this study, global minimum point is determined at
(3,2) location with PEA.

136
Test results of the Himmelblau’s test function are presented in Fig. 6.5 for initial
generation (Ng=1), Ng=3, Ng=7 and Ng=10 generation number by PEA. Test results
show that PEA can find and converge the minimum point for Himmelblau’s test
function. Global minimum point is found in (3, 2.003) location by PEA for
Himmelblau’s test function.

Figure 6.5 Solutions of the Himmelblau’s test function by PEA with 10 generation

6.6 Perturbation-Based Evolutionary Algorithm with Beale Test Function

The test function of Beale is a multimodal function with four sharp peak points at
the corner which has global minimum at (3,0.5) and one local maximum. The search
space bounds, and graphical illustration of the Beale test function are shown in the
Table 6.2 in its two-dimensional form. In Beale test function, global minimum point
is located in (3,0.5) point.

137
Test results of the Beale test function are presented in Fig. 6.6 for initial generation
(Ng=1), Ng=3, Ng=7 and Ng=10 generation number by PEA. Test results show that
PEA can find the global minimum point in (3.009,0.502) location and converge the
minimum point in 10. generation for Beale test function.

Figure 6.6 Solutions of the Beale test function by PEA with 10 generation

138
CHAPTER SEVEN
CONCLUSION

Inversion studies related with the geophysical synthetic and field data sets are
presented using two metaheuristic methods named Cuckoo Search Algorithm and Bat
Algorithm which have not been applied on the geophysical data for the inversion. Each
metaheuristic method is used for the inversion after the determination of the control
parameters of the algorithm by parameter tuning study. In parameter tuning study, each
combination of the control parameters is used for the inversions to determine the
standard deviation and min, max and mean rms.

According to the determined control parameters of the CSA, the inversion results
of the SP anomaly using CSA can be defined as follows:

• Estimated model parameters of synthetic without noise and noise-contaminated


SP data sets from the inversion with CSA are found close compared to the true
values of the model parameters. Higher difference between true value and
estimated values is observed at the parameter K for both without noise and noise-
contaminated data. Inversion results show that 0.063 mV and 2.042 mV rms
values are obtained for without noise and noise-contaminated data respectively.
Because of the ±2 mV random noise addition to create noisy data, 2.042 mV rms
can be described as low and it shows the good agreement of the model
parameters and high performance of the CSA for the model parameter
estimations of synthetic data sets.
• Changing plots of the model parameters depending on the iterations number for
without noise and noise-contaminated SP data indicate that all model parameters
are converge their best results after the 200th iteration. These results show that
the good convergence results of the CSA on the inversions of synthetic noise
free and noisy SP data.
• It is seen that CSA is the useful option for the inversion of the multiple synthetic
SP anomalies. Inversion results shows good agreement at observed and

139
calculated anomaly for two sourced multiple SP anomaly with 1.76 mV rms
result. Model parameter estimations with CSA is extremely high to the true
values especially at the second source of the anomaly.
• For the inversion of the field data sets, CSA is used for the inversion of four field
data which one of them is multiple field data anomaly occurred by two sources.
For Bavarian woods anomaly, 10.95 mV rms is obtained by presented study
which is the lowest rms values compared by the literature. Inversion of
Süleymanköy and Malachite mine anomaly by CSA yield 12.19 mV and 6.62
mV rms respectively, which are similar with the literature. On the other hand,
CSA is applied on a multiple field data called KTB Borehole anomaly and good
agreement is yielded between observed and calculated anomaly.

After the inversions of the SP anomalies by CSA, the performance of the CSA is
evaluated by the inversion of the gravity data sets. CSA is discussed for the inversion
of the gravity as follows:

• CSA is applied on the gravity noise free and noisy data sets for the model
parameter estimation and the results are very close especially for the noise free
data sets. As a result of the CSA estimations with high accuracy, rem values are
obtained 0.0011 mGal and 0.256 mGal for without noise and noise-contaminated
data respectively. In noisy data, these rms result for the noisy data is yielded
noise addition with ±0.25 mGal standard deviation. Therefore, it can be said that
CSA cause low rms results in the inversion of the synthetic gravity anomalies
with and without noise.
• High convergence of the model parameters according to the iterations at
synthetic gravity anomalies are observed in z0 and x0 which are converged after
the iteration number 150. On the other hand, model parameters A, q and η are
converged at 200th iterations.
• As field anomalies, CSA is used for the inversion of the two field anomalies
called Camaguey and Quebec anomaly. In Camaguey anomaly, 0.00431 mGal
rms result is found with good agreement of the observed and calculated anomaly
and low error energy. However, convergence of the model parameters are not
good enough as other anomalies. On the other hand, low rms as 0.033 mGal, low

140
error energy and better convergence of the model parameters are observed in
Quebec anomaly. It is clearly seen that model parameters are converged after the
iteration number 50 for the run with 300 iterations. This can be a great example
to show the convergence performance of the CSA on the field anomaly.

In addition, the results of the inversion of SP and gravity anomalies using CSA are
evaluated for the uncertainty appraisal analysis by M-H method and feasible relative
frequency distributions with low standard deviations are observed for the all-model
parameters of the synthetic and field SP and gravity anomalies.

The inversion of the SP and gravity anomalies are carried out using BA as the
second metaheuristic method for the inversion studies in this thesis. First, the control
parameters of the BA are determined by the parameter tuning study. Then, the
inversions are performed using BA and the results are evaluated as follows:

• Model parameter estimations of SP anomalies are achieved by the inversions


using BA with 0.098 and 2.040 mV rms for noise free and noisy synthetic data
respectively. According to the results of the anomaly, observed and calculated
values fit with low error energy. Also, model parameters are converged after the
iteration number 150 for the run with 300 iteration using BA to invert noise free
and noisy SP anomalies. Additionally, error energy plot of the noisy SP anomaly
is converged at 100th iteration with BA. It can be understood that the BA is a
successful metaheuristic method for the inversion of the noise-contaminated
anomalies.
• BA yields good relative frequency distribution at the inversion of without noise
and noise-contaminated SP anomalies with low standard deviation. Uncertainty
analysis of M-H samples show that mean values are close to the estimated values
of the model parameters using BA which is the reason of low standard deviation
and feasible distribution.
• For multiple synthetic SP anomalies, the inversion results give 3.62 mV rms with
BA. It is seen that model parameters of Source-2 of the anomaly are close to the
true value, but one parameter of the Source-1 is a little far from its true value for
the multiple synthetic SP anomaly.

141
• BA is used for the inversion of the SP field anomalies and similar rms results are
obtained with CSA for all field anomalies. In Bavarian, Malachite and
Süleymanköy anomalies, error energy is converged before the 100th iteration of
the run with BA and 10.94 mV, 6.558 mV and 12.19 mV rms results are found
respectively. According to the literature investigation for the field anomalies
with various metaheuristic methods, BA gives lower results for Bavarian and
Malachite anomalies than the results taken from the literature. Additionally, the
inversion results of the KTB Borehole anomaly show that BA is successful to
find the model parameters of the multiple SP field anomaly.

Evaluations of the inversion results of the synthetic and field gravity anomalies
using BA are summarized as follows:

• The inversions of the without noise and noise-contaminated gravity anomalies


give low rms results as 0.001 mGal and 0.256 mGal respectively. On the other
hand, convergence of the model parameters is obtained at 250th and 150th
iteration for without noise and noise-contaminated gravity anomalies
respectively. In noisy anomaly, error energy reaches its minimum value at 50th
iteration with BA.
• Distribution of relative frequency of the model parameters for without noise and
noise-contaminated gravity anomalies are obtained with low standard deviation
by M-H uncertainty analysis.

Inversion results of the gravity field anomalies using BA show that Camaguey and
Quebec anomaly gives low rms values similar to the results that studied in the
literature. Error energy and model parameter values are converged before the 100th
iteration in the inversion results with BA for Camaguey anomaly. On the other hand,
in Quebec anomaly, the convergence is achieved at 250th iteration for the model
parameters of the gravity field data.

Based on the inversion studies, it was observed that the metaheuristic algorithms
used in this thesis are able to yield compatible solutions with true values (for synthetic

142
anomalies) and the results from the literature (for field anomalies). Considering fast
and good convergence characteristics, metaheuristic algorithms of CSA and BA can
be performed as a powerful tool for parameter estimations from synthetic and field
data sets. Finally, Perturbation-Based Evolutionary Algorithm (PEA) is introduced as
a new metaheuristic algorithm for geophysical inversion studies. The efficiency of
PEA is tested successfully with 5 different test functions, and it is understood that PEA
is suitable for geophysical inversion studies.

143
REFERENCES

Abdelrahman E.M., Ammar A.A., Sharafeldin S.M. and Hassanein H.I. (1997). Shape
and depth solutions from numerical horizontal self-potential gradients. Journal of
Applied Geophysics, 37(1), 31-43.

Abdelrahman, E.M., Bayoumi, A.I., Abdelhady, Y.E., Gobashy, M.M. and El-Araby,
H.M. 1989. Gravity interpretation using correlation factors between successive
least-squares residual anomalies, Geophysics, 54(12), 1614-1621.

Abdelazeem, M. and Gobashy, M. (2006). Self-Potential Inversion Using Genetic


Algorithm. JKAU: Earth Science, 17, 83-101.

Abdelazeem, M., Gobashy, M., Khalil, M.H. and Abdrabou, M. (2019). A complete
model parameter optimization from self-potential data using Whale algorithm.
Journal of Applied Geophysics, 170, 103825.

Abdmouleh, Z., Gastli, A., Ben-Brahim, L., Haouari, M. and Al-Emadi, N.A. (2017).
Review of optimization techniques applied for the integration of distributed
generation from renewable energy sources. Renewable Energy, 113, 266-280.

Agarwal, A., Chandra, A., Shalivahan, S. and Singh, R. (2018). Grey wolf optimizer:
a new strategy to invert geophysical data sets. Geophysical Prospecting, 66, 1215-
1226.

Akca, I., Günther, T., Müller-Petke, M., Başokur, A.T. and Yaramancı, U. (2014).
Joint parameter estimation from magnetic resonance and vertical electric
soundings using a multi-objective genetic algorithm. Geophysical Prospecting,
62, 364-376.

144
Aleardi, M. (2015). Seismic velocity estimation from well log data with genetic
algorithms in comparison to neural networks and multilinear approaches. Journal
of Applied Geophysics, 117, 13-22.

Alkan, H. and Balkaya, Ç. (2018). Parameter estimation by Differential Search


Algorithm from horizontal loop electromagnetic (HLEM) data. Journal of Applied
Geophysics, 149, 77-94.

Alvandi, A. and Asil, R.H. (2018). A new approach for residual gravity anomalies
ınterpretations: artificial bee colony optimization algorithm. The Journal of Indian
Geophysical Union, 22(1), 24-31.

Amjadi, A. and Naji, J. (2013). Application of Genetic Algorithm Optimization and


Least Square Method for depth determination from residual gravity anomalies.
International Research Journal of Applied and Basic Sciences, 5(5), 661-666.

Anderson, N.L., Essa, K.S. and Elhussein, M. (2020). A comparison study using
particle swarm optimization inversion algorithm for gravity anomaly
interpretation due to a 2D vertical fault structure. Journal of Applied Geophysics,
179, 104120.

Attwa, M., Gemail, K., Eleraki, M. and Zamzam, S. (2014). Application of genetic
algorithm (GA) for soil characterization using DC resistivity data: A case study in
Tenth of Ramadan city, Egypt. Advances in Natural and Applied Sciences, 8(1),
38-50.

Ayani, M., Mallick, S., Hunziker, J. and MacGregor, L. (2017). Inversion of frequency
domain marine controlled-source electromagnetic data using genetic algorithm.
SEG Technical Program Expanded Abstracts, 1252-1256.

145
Ayani, M., MacGregor, L. and Mallick, S. (2020). Inversion of marine controlled
source electromagnetic data using a parallel non-dominated sorting genetic
algorithm. Geophysical Journal International, 220, 1066-1077.

Balkaya Ç. (2013). An implementation of differential evolution algorithm for


inversion of geoelectrical data. Journal of Applied Geophysics, 98, 160-175.

Balkaya, Ç., Ekinci, Y.L., Göktürkler, G. and Turan, S. (2017). 3D non-linear


inversion of magnetic anomalies caused by prismatic bodies using differential
evolution algorithm. Journal of Applied Geophysics, 136, 372-386.

Balkaya, Ç. and Kaftan, I. (2021). Inverse modelling via differential search algorithm
for interpreting magnetic anomalies caused by 2D dyke-shaped bodies. Journal of
Earth System Science, 130, 135.

Beaty, K.S., Schmitt, D.R. and Sacchi, M. (2002). Simulated annealing inversion of
multimode Rayleigh wave dispersion curves for geological structure. Geophysical
Journal International, 151, 622-631.

Ben, U.C., Akpan, A.E., Mbonu, C.G. and Ebong, E.D. (2021). Novel methodology
for interpretation of magnetic anomalies due to two-dimensional dipping dikes
using the Manta Ray Foraging Optimization. Journal of Applied Geophysics, 192,
104405.

Ben, U.C., Akpan, A.E., Mbonu, C.G., Urang, J.G., Akaerue, E.I. and Obianwu, V.I.
(2022). Novel methodology for the geophysical interpretation of magnetic
anomalies due to simple geometrical bodies using social spider optimization
(SSO) algorithm. Heliyon, 8, e09027.

Ben, U.C., Ekwok, S.E., Akpan, A.E., Mbonu, C.G., Eldosouky, A.M., Abdelrahman,
K. and Gomez-Ortiz, D. (2022). Interpretation of Magnetic Anomalies by Simple

146
Geometrical Structures Using the Manta-Ray Foraging Optimization. Frontiers in
Earth Science, 10, 849079.

Biswas A. (2017). A review on modeling, inversion and interpretation of self-potential


in mineral exploration and tracing paleo-shear zones. Ore Geology Reviews 91,
21-56.

Biswas, A. (2015). Interpretation of residual gravity anomaly caused by simple shaped


bodies using very fast simulated annealing global optimization. Geoscience
Frontiers, 6, 875-893.

Biswas, A. (2016a). Interpretation of gravity and magnetic anomaly over thin sheet-
type structure using very fast simulated annealing global optimization technique.
Modeling Earth Systems and Environment, 2(30), 1-12.

Biswas, A. (2016b). A Comparative Performance of Least-Square Method and Very


Fast Simulated Annealing Global Optimization Method for Interpretation of Self-
Potential Anomaly over 2-D Inclined Sheet Type Structure. Journal Geological
Society of India, 88, 493-502.

Biswas, A. and Acharya, T. (2016). A very fast simulated annealing method for
inversion of magnetic anomaly over semi-infinite vertical rod-type structure.
Modeling Earth Systems and Environment, 2(198), 1-10.

Biswas, A. and Sharma, S.P. (2014). Optimization of self-potential interpretation of 2-


D inclined sheet-type structures based on very fast simulated annealing and
analysis of ambiguity. Near Surface Geophysics, 13, 179-195.

Biswas, A. and Sharma, S.P. (2015). Interpretation of self-potential anomaly over


idealized bodies and analysis of ambiguity using very fast simulated annealing
global optimization technique. Journal of Applied Geophysics, 105, 235-247.

147
Biswas, A. and Sharma, S.P. (2017). Interpretation of self-potential anomaly over 2-D
inclined thick sheet structures and analysis of uncertainty using very fast simulated
annealing global optimization. Acta Geodaetica et Geophysica, 52, 439-455.

Blum, C. and Roli, A. (2003) Metaheuristics in Combinatorial Optimization:


Overview and Conceptual Comparison”, ACM Computing Surveys, 35(3), 268-
308.

Bogoslovsky V.A. and Ogilvy A.A. (1977). Geophysical methods for the investigation
of landslides. Geophysics 42(3), 562-571.

Bouchaoui, L., Ferahtia, J., Farfour, M. and Djarfour, N. (2022 Vertical electrical
sounding data inversion using continuous ant colony optimization algorithm: A
case study from Hassi R’Mel, Algeria. Near Surface Geophysics, 20, 419-439.

Chandra, A. Agarwal, A., Shalivahan, S. and Singh, R.K. (2017). Grey wolf
optimisation for inversion of layered earth geophysical datasets. Near Surface
Geophysics, 15, 499-513.

Chauhan, M.S. (2017). MHODE inversion of potential fields. PhD thesis, Unıversità
Degli Studı Di Napoli Federico II.

Chen, H.G., Li, J.X., Wu, J.S. and Yu, P. (2012). Study on simulated-annealing MT-
gravity joint inversion. Chinese Journal of Geophysics, 55(2), 663-670.

Cheng-Yu, S., Yan-Yan, W. and Xiao-Jun, Q. (2017). Nonlinear Rayleigh wave


inversion based on the shuffled frog-leaping algorithm. Applied Geophysics,
14(4), 551-558.

Chunduru, R.K., Sen, M.K. and Stoffa, P.L. (1996). 2-D resistivity inversion using
spline parameterization and simulated annealing. Geophysics, 61(1), 151-161.

148
Conti, C.R., Roisenberg, M., Neto, G.S. and Porsani, M.J. (2013). Fast seismic
inversion methods using ant colony optimization algorithm. IEEE Geoscience and
Remote Sensing Letters, 10(5), 1119-1123.

Cui, Y., Chen, Z., Zhu, X., Liu, H. and Liu, J. (2017). Sequential and Simultaneous
Joint Inversion of Resistivity and IP Sounding Data Using Particle Swarm
Optimization. Journal of Earth Science, 28(4), 709-718.

Cui, Y., Zhang, L., Zhu, X., Liu, J. and Guo, Z. (2020). Inversion for magnetotelluric
data using the particle swarm optimization and regularized least squares. Journal
of Applied Geophysics, 181, 104156.

Currenti, G., Negro, C.D., Fortuna, L., and Ganci, G. (2007). Integrated inversion of
ground deformation and magnetic data at Etna volcano using a genetic algorithm
technique. The 3rd International Conference on Natural and Environmental
Sciences (ICONES), 364, 012035.

Das, S. and Suganthan, P.N. 2010. Differential evolution: A survey of the state-of-the-
art. IEEE transactions on evolutionary computation, 15(1), 4-31.

Darisma, D. and Marwan (2019). One-dimensional magnetotelluric inversion using


levenberg-marquardt and particle swarm optimization algorithm. Annals of
Geophysics, 50(1), 21-30.

Davis, W.E., Jackson, W.H. and Richter, D.H. (1957). Gravity prospecting for
chromite deposits in Camaguey province, Cuba. Geophysics, 22(4), 848-869.

Di Maio, R., Milano, L. and Piegari, E. (2020). Modeling of magnetic anomalies


generated by simple geological structures through Genetic-Price inversion
algorithm. Physics of the Earth and Planetary Interiors, 305, 106520.

149
Di Maio R., Piegari E. and Rani P. (2017). Source depth estimation of self-potential
anomalies by spectral methods. Journal of Applied Geophysics, 136, 315-325.

Di Maio, R., Piegari, E., Rani, P, Carbonari, R., Vitagliano, E. and Milano, L. (2019).
Quantitative interpretation of multiple self-potential anomaly sources by a global
optimization approach. Journal of Applied Geophysics, 162, 152-163.

Di Maio, R., Rani, P., Piegari, and E. Milano, L. (2016). Self- potential data inversion
through a Genetic-Price algorithm. Computers&Geosciences, 94, 86-95.

Dittmer, J.K. and Szymanski, J.E. (1995). The stochastic inversion of magnetics and
resistivity data using the simulated annealing algorithm. Geophysical Prospecting,
43, 397-416.

Dixon, L.C.W. and Szego, G.P. (1978). The global optimization problem: an
introduction. Towards Global Optimization, 2, 1-15.

Dong, L., Li, D. and Jiang, F. (2015). The DE algorithm and its application for IP
effect actraction of magnetotelluric sounding data. Progress in Geophysics, 30(4),
1882-1895.

Dorigo, M. (1992). Optimization, Learning and Natural Algorithms (in Italian). PhD
thesis, Dipartimento di Elettronica, Politecnico di Milano, Milan, Italy.

Ekinci, Y.L., Balkaya, Ç. and Göktürkler, G. (2019). Parameter estimations from


gravity and magnetic anomalies due to deep-seated faults: differential evolution
versus particle swarm optimization. Turkish Journal of Earth Sciences, 28, 860-
881.

Ekinci Y.L., Balkaya Ç., Göktürkler G. 2020. Global Optimization of Near-Surface


Potential Field Anomalies Through Metaheuristics. In: Biswas A., Sharma S.

150
(eds) Advances in Modeling and Interpretation in Near Surface Geophysics.
Springer Geophysics. Springer.

Ekinci, Y.L., Balkaya, Ç., Göktürkler, G. and Özyalın, Ş. (2021). Gravity data
inversion for the basement relief delineation through global optimization: a case
study from the Aegean Graben System, western Anatolia, Turkey. Geophysical
Journal International, 224, 923-944.

Ekinci, Y.L., Balkaya, Ç., Göktürkler, G. and Turan, S. (2016). Model parameter
estimations from residual gravity anomalies due to simple-shaped sources using
Differential Evolution Algorithm. Journal of Applied Geophysics, 129, 133-147.

Ekinci, Y.L., Özyalın, Ş., Sındırgı, P., Balkaya, Ç. and Göktürkler, G. (2017).
Amplitude inversion of the 2D analytic signal of magnetic anomalies through the
differential evolution algorithm. Journal of Geophysics and Engineering, 14,
1492-1508.

El-Araby H.M. (2004). A new method for complete quantitative interpretation of self-
potential anomalies. Journal of Applied Geophysics, 55, 211–224.

Essa, K.S. (2019). A particle swarm optimization method for interpreting self-potential
anomalies. Journal of Geophysics and Engineering, 16, 463-477.

Essa, K.S. (2020). Self potential data interpretation utilizing the particle swarm method
for the finite 2D inclined dike: mineralized zones delineation. Acta Geodaetica et
Geophysica, 55, 203-221.

Essa, K.S. and Elhussein, M. (2018). PSO (Particle Swarm Optimization) for
Interpretation of Magnetic Anomalies Caused by Simple Geometrical Structures.
Pure and Applied Geophysics, 175, 3539-3553.

151
Essa, K.S. and Elhussein, M. (2020). Interpretation of Magnetic Data Through Particle
Swarm Optimization: Mineral Exploration Cases Studies. Natural Resources
Research, 29(1), 521-537.

Essa, K.S. and Geraud, Y. (2020). Parameters estimation from the gravity anomaly
caused by the two-dimensional horizontal thin sheet applying the global particle
swarm algorithm. Journal of Petroleum Science and Engineering, 193, 1077421.

Essa, K.S., Mehanee, S.A. and Elhussein, M. (2021). Gravity data interpretation by a
two-sided fault-like geologic structure using the global particle swarm technique.
Physics of the Earth and Planetary Interiors, 311, 106631.

Essa, K.S. and Munschy, M. (2019). Gravity data interpretation using the particle
swarm optimization method with application to mineral exploration. Journal of
Earth System Science, 128:123, 1-16.

Fenton, M.B. (2004). Bat natural history and echolocation. In: Brigham RM, Elisabeth
VKV, Gareth, J., Stuart, P. and Herman, A.L. (eds.), Bat echolocation research
tools, rechniques and analysis (2-6). Bat Conservation International, Austin,
Texas, USA.

Franke, W. (1989). The geological framework of the KTB drill site, Oberpfalz. In R.
Emmermann & J. Wohlenberg (Eds.), The German continental drilling program
(KTB), New York: Springer, pp. 37–54.

Franke, W. (1989). The geological framework of the KTB drill site, Oberpfalz, in The
German Continental Deep Drilling Program (KTB), edited by R. Emmermann and
J. Wohlenberg, pp. 37-54, Springer-Verlag, New York.

Galassi M., Davies J., Theiler J., Gough B.. Jungman G., Alken P., Booth M. and Rossi
F. (2009). GNU Scientific Library Reference Manual 3rd edn, Bristol: Network
Theory Ltd.

152
Gandomi, A.H., Yang, X.S. and Alavi, A.H. (2013) Cuckoo search algorithm: a
metaheuristic approach to solve structural optimization problems. Engineering
Computations, 29(1), 17–35.

Gobashy, M., Abdelazeem, M., Abdrabou, M. and Khalil, M.H. (2020). Estimating
model parameters from self-potential anomaly of 2D inclined sheet using whale
optimization algorithm: applications to mineral exploration and tracing shear
zones. Natural Resources Research, 29(1), 499-519.

Godio, A. and Santilano, A. (2018). On the optimization of electromagnetic


geophysical data: Application of the PSO algorithm. Journal of Applied
Geophysics, 148, 163-174.

Gogna, A. and Tayal, A. (2013). Metaheuristics: review and application. Journal of


Experimental & Theoretical Artificial Intelligence, 25(4), 503-526.

Goldberg, D. E. 1989a. Genetic Algorithms, Book Chapter: Computer Implementation


of a Genetic Algorithm. Addison-Wesley Publishing Company, Inc, 10, 85-86.

Goldberg, D. E. 1989b. Genetic algorithms in search, optimization, and machine


learning. Addison-Wesley.

Göktürkler, G. 2011. A hybrid approach for tomographic inversion of crosshole


seismic first-arrival times. Journal of Geophysics and Engineering, 8, 99-108.

Göktürkler, G. and Balkaya, Ç. (2012). Inversion of self-potential anomalies caused


by simple-geometry bodies using global optimization algorithms. Journal of
Geophysıcs and Engıneerıng, 9, 498-507.

Göktürkler G., Balkaya Ç., Erhan Z. and Yurdakul A. (2008). Investigation of a


shallow alluvial aquifer using geoelectrical methods: a case from Turkey.
Environmental Geology 54(6), 1283-1290.

153
Grandis, H. and Maulana. (2017). Particle Swarm Optimization (PSO) for
Magnetotelluric (MT) 1D Inversion Modeling. Southeast Asian Conference on
Geophysics, 62, 012033.

Grant, F.S., West, G.F. (1965). Interpretation Theory in Applied Geophysics,


McGraw-Hill Book Company, 583.

Guo, C., Yang, Z., Wu, X., Tan, T. and Zhao, K. (2019). Application of an Adaptive
Multi-Population Parallel Genetic Algorithm with Constraints in Electromagnetic
Tomography with Incomplete Projections. Applied Science, 9, 2611.

Haryono, A., Sungkono, S., Agustin, R., Santosa, B.J., Widodo, A. and Ramadhany,
B. (2020). Model parameter estimation and its uncertainty for 2‑D inclined sheet
structure in self‑potential data using crow search algorithm. Acta Geodaetica et
Geophysica, 55, 691-715.

Hastings W. (1970). Monte Carlo sampling methods using Markov chains and their
applications. Biometrika, 57(1), 97-109.

Haupt, R.L. and Haupt, S.E. 2004. Practical Genetic Algorithms. John Wiley & Sons,
Inc, New Jersey, USA.

Heidari, M., Meshinchi Asl, M., Mehramuz, M and Heidari, R. 2022. Magnetic field
analysis using the improved global particle swarm optimization algorithm to
estimate the depth and approximate shape of the buried mass. Contributions to
Geophysics and Geodesy, 52/2, 281-305.

Heiland C.A., Tripp R.M. and Dart W. (1945). Geophysical surveys at the Malachite
Mine, Jefferson County, Colorado. Transactions of the American Institute of
Mining, Metallurgical and Petroleum Engineers, 164, 142-154.

Holland, J. H. 1975. An efficient genetic algorithm for the traveling salesman problem.

154
European Journal of Operational Research, 145, 606-617.

Jamasb, A., Motavalli-Anbaran, S.H. and Ghasemi, K. (2019). A Novel Hybrid


Algorithm of Particle Swarm Optimization and Evolution Strategies for
Geophysical Non-linear Inverse Problems. Pure and Applied Geophysics, 176,
1601-1613.

Jie, X. and Tao, Z. (2015). Multiobjective particle swarm inversion algorithm for two-
dimensional magnetic data. Applied Geophysics, 12(2), 127-136.

Joolaei, A., Amiri, A.A. and Nejati, A. (2021). Gravity inversion of basement relief
using imperialist competitive algorithm with hybrid techniques. Acta Geophysica,
1-18.

Kaftan, I. (2017). Interpretation of magnetic anomalies using a genetic algorithm. Acta


Geophysica, 65, 627-634.

Karaboğa, D. and Ökdem, S. 2004. A simple and global optimization algorithm for
engineering problems: differential evolution algorithm. Turkish Journal of
Electrical Engineering & Computer Sciences, 12(1), 53-60.

Karcıoğlu, G. and Gürer, A. (2017). Implementation and model uniqueness of Particle


Swarm Optimization method with a 2D smooth modeling approach for Radio-
Magnetotelluric data. Journal of Applied Geophysics, 169, 37-48.

Kearey, P., Brooks, M., and Hill, I. (2002). An Introduction to Geophysical


Exploration, Third Edition. Blackwell Science Ltd., Malden, USA.

Kennedy, J. and Eberhart, R., 1995. Particle swarm optimization. IEEE International
Conference on Neural Networks, pp. 1942–1948, Piscataway, NJ, USA

155
Kirkpatrick, S., Gerlatt, C. D., Vecchi, M.P. 1983. Optimization by Simulated
Annealing, Science, 220, 671-680.

Krüger, T.J. (2017). Development, implementation and theoretical analysis of the bee
colony optimization meta-heuristic method, PhD Thesis. University of Novi Sad,
Novi Sad, Serbia.

Kumar, V. and Gupta, P.K. (2012). Nonlinear one-dimensional seismic waveform


inversion using Harmony search. International Geophysical Conference and Oil
& Gas Exhibition, 17-19 September, Istanbul, Turkey.

Li, H., Wang, H., Wang, L. and Zhou, X. (2020). A modified Boltzmann Annealing
Differential Evolution algorithm for inversion of directional resistivity logging-
while-drilling measurements. Journal of Petroleum Science and Engineering, 188,
106916.

Li, J., Yang, J. and Chen, X. (2013). Software design of genetic algorithm for rayleigh
wave dispersion curve inversion. Progress in Geophysics, 28(5), 2693-2700.

Li, J., Yan, H., Tang, J.T., Zhang, X., Li, G. and Zhu, H. (2018). Magnetotelluric noise
suppression based on matching pursuit and genetic algorithm. Chinese Journal of
Geophysics, 61(7), 3086-3101.

Li, M., Cheng, J., Wang, P., Xiao, Y., Yao, W., Su, C., Cheng, S., Guo, J. and Yu, X.
(2019). Transient electromagnetic 1D inversion based on the PSO–DLS
combination algorithm. Exploration Geophysics, 50(5), 472-480.

Li, R., Yu, N., Li, R., Zhuang, Q. And Zhang H. (2018). Transient electromagnetic
inversion based on particle swarm optimization and differential evolution
algorithm. Near Surface Geophysics, 19, 59-71.

156
Li, R., Zhang, H., Zhuang, Q., Li, R. and Chen, Y. (2018). BP neural network and
improved differential evolution for transient electromagnetic inversion.
Computers and Geosciences, 137, 104434.

Li, S.Y., Wang, S.M., Wang, P.F., Su, X.L., Zhang, X.S. and Dong, Z.H. (2018). An
improved grey wolf optimizer algorithm for the inversion of geoelectrical data.
Acta Geophysica, 66, 607-621.

Li, X. and Yin, M. (2012). Application of differential evolution algorithm on self-


potential data. PLoS ONE, 7(12), e51199.

Liang, X., Xu, S., Liu, Y. and Sun, L. (2022). A modified whale optimization
algorithm and its application in seismic inversion problem. Mobile Information
Systems, 2022, 9159130.

Liu, B., Li, S.C., Nie, L.C., Wang, J., L, X. and Zhang, Q.S. (2012). 3D resistivity
inversion using an improved Genetic Algorithm based on control method of
mutation direction. Journal of Applied Geophysics, 87, 1-8.

Liu, S., Hu, X. and Liu, T. (2014). A stochastic inversion method for potential field
data: ant colony optimization. Pure and Applied Geophysics, 171, 1531-1555.

Louis, S.J. and Li, G. 2000. Case injected genetic algorithms for traveling salesman
problems. Information Sciences, 122(2-4), 201-225.

Ma, X.Q. (2002). Simultaneous inversion of prestack seismic data for rock properties
using simulated annealing. Geophysics, 67(6), 1877-1885.

Ma J., Ting T.O., Man K.L., Zhang N., Guan S. and Wong P.W.H. (2013). Parameter
Estimation of Photovoltaic Models via Cuckoo Search, Journal of Applied
Mathematics, 2013, 362619.

157
Madan, K.J., Kumar, S. and Chowdhury, A. (2008). Vertical electrical sounding
survey and resistivity inversion using genetic algorithm optimization technique.
Journal of Hydrology, 359, 71-87.

Mahbuby, H., Safari, A. and Foroughi, I. (2017). Local gravity field modeling using
spherical radial basis functions and a genetic algorithm. Comptes Rendus
Geoscience, 349, 106-113.

Mareli M. and Twala B. (2018). An adaptive Cuckoo search algorithm for


optimisation. Applied Computing and Informatics, 14(2), 107-115.

Martinez, J.L.F., Gonzalo, E.G., Alvarez, J.P.F., Kuzma, H.A. and Perez, C.O.M.
(2010). PSO: A powerful algorithm to solve geophysical inverse problems
Application to a 1D-DC resistivity case. Journal of Applied Geophysics, 71, 13-
25.

Mayer, D. G., Kinghorn, B.P. Archer, A. A. 2005. Differential evolution–an easy and
efficient evolutionary algorithm for model optimisation. Agricultural Systems,
83(3), 315-328.

Meiser P. (1962). A method for quantitative interpretation of self potential


measurements. Geophysical Prospecting, 10(2), 203-218.

Metropolis, N., Rosenbluth, A., Rosenbluth, M., Teller, A., Teller, E. 1953. Equations
of state calculations by fast computing machines, Journal of Chemical Physics,
21(6), 1087-1092.

Montesinon, F.G., Arnoso, J. and Vieira, R. (2005). Using a genetic algorithm for 3-
D inversion of gravity data in Fuerteventura (Canary Islands). International
Journal of Earth Science (Geol Rundsch), 94, 301-316.

158
Montesinon, F.G., Montanegro, I.B. and Arnoso, J. (2016). Three-dimensional inverse
modelling of magnetic anomaly sources based on a genetic algorithm. Physics of
the Earth and Planetary Interiors, 253, 74-87.

Moorkamp, M., Jones, A.G. and Eaton, D.W. (2007). Joint inversion of teleseismic
receiver functions and magnetotelluric data using a genetic algorithm: Are seismic
velocities and electrical conductivities compatible?. Geophysıcal Research
Letters, 34, 1-5.

Mosegaard, K. and Vestergaard, P.D. (1991). A simulated annealing approach to


seismic model optimization with sparse prior information. Geophysıcal
Prospecting, 39, 599-611.

Mundim, K.C., Lemaire, T.J. and Bassrei, A. (1998). Optimization of non-linear


gravity models through generalized simulated annealing. Physica A, 252, 405-416.

Nagihara, S. and Hall, S.A. (2001). Three-dimensional gravity inversion using


simulated annealing: Constraints on the diapiric roots of allochthonous salt
structures. Geophysics, 66(5), 1438-1449.

Nguyen T.T., Dieu N.V., Truong A.V. and Ho L.D. (2016). Meta-Heuristic
Algorithms for Solving Hydrothermal System Scheduling Problem Considering
Constraints in Transmission Lines. Global Journal of Technology and
Optimization, 7, 1-6.

Ni, R. and Liang, X (2022). Seismic Inversion Problem Using a Multioperator Whale
Optimization Algorithm. Discrete Dynamics in Nature and Society, 2022,
1966054.

Osman, O. and Albora, A.M. (2015). Modeling of gravity anomalies due to 2d


geological structures using genetic algorithm. Electrica, 15(2), 1929-1935.

159
Pace, F., Santilano, A. and Godio, A. (2019). Particle Swarm Optimization of 2D
Magnetotelluric data. Geophysics, 84(3), 1-59.

Pallero, J.L.G., Fernandez-Martinez, J.L., Bonvalot, S. and Fudym, O. (2017). 3D


gravity inversion and uncertainty assessment of basement relief via Particle
Swarm Optimization. Journal of Applied Geophysics, 139, 338-350.

Pallero, J.L.G., Fernandez-Martinez, J.L., Fernandez-Muniz, Z., Bonvalot, S.,


Gabalda, G. and Nalpas, T. (2021). GRAVPSO2D: A Matlab package for 2D
gravity inversion in sedimentary basins using the Particle Swarm Optimization
algorithm. Computers and Geosciences, 146, 104663.

Pan, Z., Liang, H., Gao, Z. and Gao, J. (2018). Differential evolution with
subpopulations for high-dimensional seismic inversion. Geophysical Prospecting,
66(6), 1060-1069.

Parasnis, D.S. (1986). Principles of Applied Geophysics, Fourth Edition, Chapman


and Hall, New York.

Patel, S.K., Pandey, A.K., Roshan, R. and Singh, U.K. (2016). Application of PSO and
GSA Hybrid Optimization Method for 1-D Inversion of Magnetotelluric Data.
International conference on Signal Processing, Communication, Power and
Embedded System (SCOPES).

Pekşen, E., Yas, T. and Kıyak, A. (2014). 1-D DC Resistivity Modeling and
Interpretation in Anisotropic Media Using Particle Swarm Optimization. Pure and
Applied Geophysics, 171, 2371-2389.

Pekşen, E., Yas, T., Kayman, A.Y. and Özkan, C. (2011). 1 Application of particle
swarm optimization on self-potential data. Journal of Applied Geophysics, 75,
305-318.

160
Pendock, N. (1993). A genetic algorithm for conductivity imaging of airborne
electromagnetic data. Proc. SPIE, 1942, 129-136.

Peres, F. and Castelli, M. (2021). Combinatorial Optimization Problems and


Metaheuristics: Review, Challenges, Design, and Development. Applied Science,
11, 6449.

Poormirzaee, R. (2016). S-wave velocity profiling from refraction microtremor


Rayleigh wave dispersion curves via PSO inversion algorithm. Arabian Journal
of Geosciences, 9, 673.

Poormirzaee, R. (2017). Applying bat metaheuristic algorithm for building shear wave
velocity models from surface wave dispersion curves. Conference Proceedings,
23rd European Meeting of Environmental and Engineering Geophysics,
September, 1-5.

Poormirzaee, R., Moghadam, R.H. and Zarean, A. (2014). Introducing Particle Swarm
Optimization (PSO) to Invert Refraction Seismic Data. 20th European Meeting of
Environmental and Engineering Geophysics, 14-18 December, Athens, Greece.

Poormirzaee, R., Moghadam, R.H. and Zarean, A. (2015). Inversion seismic refraction
data using particle swarm optimization: a case study of Tabriz, Iran. Arabian
Journal of Geosciences, 8, 5981-5989.

Poormirzaee, R., Nesa, E., Ramin, N. And Mohammad, N.B. (2020). Inversion of
Rayleigh wave dispersion curves using modified flower pollination algorithm
(MFPA) to detect shear wave velocity profile. Journal of Research on Applied
Geophysics, 6(1), 37-49.

Poormirzaee, R., Sarmady, S. and Sharghi, Y. (2019). A new inversion method using
a modified bat algorithm for analysis of seismic refraction data in Dam site
investigation. Journal of Environmental and Engineering Geophysics, 24(2), 201.

161
Qing, A., Lee, C.K. and Jen, L. (2001). Electromagnetic Inverse Scattering of Two-
Dimensional Perfectly Conducting Objects by Real-Coded Genetic Algorithm.
IEEE Transactıons on Geoscıence and Remote Sensıng, 39(3), 665-676.

Qiu, N., Liu, Q. and Gao, Q. (2009). Gravity Data Inversion Based Genetic Algorithm
and Generalized Least Squares. IEEE International Conference on Intelligent
Computing and Intelligent Systems, 891-893.

Qiu, T., Zhao, Y., Hu, S., An, C., Bi, W., Ge, S., Capineri, L. and Bohlen, T. (2020).
An Interactive Integrated Interpretation of GPR and Rayleigh Wave Data Based
on the Genetic Algorithm. Surveys in Geophysics, 41, 549-574.

Ramadhany, B., Sungkono, S. and Rahmawati, A. (2019). Groundwater identification


using VES With Regressive- Regressive PSO Inversion in Pacitan Region,
Indonesia. 2nd Asia Pacific Meeting on Near Surface Geoscience & Engineering,
April 22-26, Kuala Lumpur, Malaysia.

Riberio, J., Okita, N.T., Coimbra, T.A. Ignacio, G.B. and Tygel, M. (2019). Using
Adaptive Differential Evolution algorithm to improve parameter estimation in
seismic processing. Sixteenth International Congress of The Brazilian
Geophysical Society, August 19-22, Rio de Janeiro, Brazil.

Richter, M. (2016). Inverse Problems Basics, Theory and Applications in Geophysics,


Birkhauser Springer International Publishing, Cham, Switzerland.

Roy, A., Dubey, C.P. and Prasad, M. (2021). Gravity inversion of basement relief
using Particle Swarm Optimization by automated parameter selection of Fourier
coefficients. Computers and Geosciences, 156, 104875.

Roy, L., Sen M.K., Blankenship, D.D., Stoffa, P.L. and Richter, T.G. (2005). Inversion
and uncertainty estimation of gravity data using simulated annealing: An
application over Lake Vostok, East Antarctica. Geophysics, 70(1), 1-12.

162
Rubaiyn, A., Priyono, A. and Yudistira, T. (2019). Near-surface S-wave Estimation
base on Inversion of Rayleigh Wave Dispersion Curve Using Genetic Algorithm.
IOP Conf. Series: Earth and Environmental Science, 318, 012013.

Salgotra R., Singh U. and Saha S. (2018). New cuckoo search algorithms with
enhanced exploration and exploitation properties. Expert Systems with
Applications, 95, 384-420.

Santos, F.A.M. (2010). Inversion of self-potential of idealized bodies’ anomalies using


particle swarm optimization. Computers&Geosciences, 36, 1185-1190.

Satyanarayana Murty, B.V. and Haricharan, P. (1985). Nomogram for the complete
interpretation of spontaneous potential profiles over sheet-like and cylindrical
two-dimensional sources. Geophysics, 50(7), 1127-1135.

Schwarzbach, C., Börner, R.U. and Spitzer, K. (2011). Two-dimensional inversion of


direct current resistivity data using a parallel, multi-objective genetic algorithm.
Geophysical Journal International, 162(3), 685-695.

Sen, M.K. and Stoffa, P.L. (1996). Bayesian inference, Gibbs' sampler and uncertainty
estimation in geophysical inversion. Geophysical Prospecting, 44(2), 313-350.

Sen, M.K. and Stoffa, P.L. (1991). Nonlinear one-dimensional seismic waveform
inversion using simulated annealing. Geophysics, 56(10), 1624-1638.

Shadmehri, M.M., Sharifi, M.A., Ardestani, V.E., Safari, A.R. and Baghani, A. (2015).
Inversion of Gravity Data Using Ant Colony Algorithm (Case Study: Gotvand-
Iran). Journal Of Geomatics Science and Technology, 4(4), 193-208.

Sharma, S.P. (2012). VFSARES—a very fast simulated annealing FORTRAN


program for interpretation of 1-D DC resistivity sounding data from various
electrode arrays. Computers &Geosciences, 42, 177-188.

163
Sharma, S.P. and Biswas, A. (2013). Interpretation of self-potential anomaly over a
2D inclined structure using very fast simulated-annealing global optimization —
An insight about ambiguity. Geophysics, 78(3), 3-15.

Sharma, S.P. and Kaikkonen, P. (1999). Global Optimisation of Time Domain


Electromagnetic Data Using Very Fast Simulated Annealing. Pure and Applied
Geophysics, 155, 149-168.

Shi, X.M., Xiao, M., Fan, J.K., Yang, G.S. and Zhang, X.H. (2009). The damped PSO
algorithm and its application for magnetotelluric sounding data inversion. Chinese
Journal of Geophysics, 52(4), 1114-1120.

Singh, B.B., Srivardhan, V., Pal, S.K., Kanagaraju, S.K., Kumar, S. and Vaish, J.
(2015). Particle Swarm Optimization Inversion of Self Potential Anomaly for
Detecting Coal Fires, a Case Study - Jharia Coal Field. Conference Proceedings,
The Third Sustainable Earth Sciences Conference and Exhibition, 1 - 5.

Song, T., Hu, X., Du, W., Cheng, L., Xiao, T. and Li, Q. (2021). Lp-Norm Inversion
of Gravity Data Using Adaptive Differential Evolution. Applied Sciences, 11,
6485.

Song, W.Q., Gao, Y.K. and Zhu, H.W. (2013). The differential evolution inversion
method based on Bayesian theory for micro-seismic data. Chinese Journal of
Geophysics, 56(4), 1331-1339.

Song, X., Li, L., Zhang, X., Shi, X., Huang, J., Cai, J., Jin, S. and Ding, J. (2014a). An
implementation of differential search algorithm (DSA) for inversion of surface
wave data. Journal of Applied Geophysics, 111, 334-345.

Song, Z., Li, L., Zhang, X., Huang, J., Shi, X., Jin, S. and Bai, Y. (2014b). Differential
evolution algorithm for nonlinear inversion of high-frequency Rayleigh wave
dispersion curves. Journal of Applied Geophysics, 109, 47-61.

164
Soupios, P., Akca, I., Mpogiatzis, P., Basokur, A.T. and Papazachos, C. (2011).
Applications of hybrid genetic algorithms in seismic tomography. Journal of
Applied Geophysics, 75, 479-489.

Srivastava, S. and Agarwal, B.N.P. (2010). Inversion of the amplitude of the two-
dimensional analytic signal of the magnetic anomaly by the particle swarm
optimization technique. Geophysical Journal International, 182, 652-662.

Srivardhan, V., Pal, S.K., Vaish, J., Kumar, S., Bharti, A.K. and Priyam, P. (2016).
Particle swarm optimization inversion of self-potential data for depth estimation
of coal fires over East Basuria colliery, Jharia coalfield, India. Environmental
Earth Sciences, 75, 688.

Storn, R. (1996). Differential Evolution Design of an IIR-Filter. In: IEEE International


conference on Evolutionary Computation (ICEC’96), Nagoya, Japan.

Storn, R. and Price, K. 1995. Differential evolution-a simple and efficient adaptive
scheme for global optimization over continuous spaces. Technical Report TR-
95-012. International Computer Science Institute, Berkeley, USA.

Storn, R. and Price, K. 1997. Differential Evolution – a simple and efficient heuristic
for global optimization over continuous spaces. Journal of Global Optimization,
11(4), 341–359.

Sungkono, S. (2020 An efficient global optimization method for self-potential data


inversion using micro-differential evolution. Journal of Earth System Science,
148, 189-200.

Sungkono, S. and Warnana, D.D. (2018). Black hole algorithm for determining model
parameter in self-potential data. Journal of Applied Geophysics, 129, 178.

165
Sweilam, N.H., El-Metwally, K. and Abdelazeem, M. (2007). Self potential signal
inversion to simple polarized bodies using the particle swarm optimization
method: A visibility study. Journal of Applied Geophysics, 6(1), 195-208.

Tlas M. and Asfahani J. (2008). Using of the adaptive simulated annealing (ASA) for
quantitative interpretation of self potential anomalies due to simple geometrical
structures. Journal of King Abdulaziz University, Earth Sciences, 19, 99-118.

Toklu, Y.C. and Bekdas, G. (2014). Metaheurıstıcs and Engineering. Proceedings of


the 15th EU/ME Workshop, Istanbul, Turkey.

Tong, X.Z., Liu, J.X., Sun, Y., Lei, W.T. and Xu, L.H. (2009). A Hybrid Optimization
Method Based on Genetic Algorithm for Magnetotelluric Inverse Problem.
International Conference on Computational Intelligence and Software
Engineering, 1-4.

Tronicke, J., Paasche, H. and Böniger, U. (2011). Joint global of GPR and P-wave
seismic traveltimes using particle swarm optimization. International workshop on
advanced ground penetrating radar (IWAGPR).

Turan-Karaoğlan, S. and Göktürkler, G. (2021). Cuckoo Search Algorithm for model


parameter estimation from self-potential data. Journal of Applied Geophysics, 194,
104461.

Turan-Karaoğlan, S. and Göktürkler, G. (2022). Gravite Anomalilerinin Guguk Kuşu


Arama Algoritması ile Ters Çözümü. Dokuz Eylul University Faculty of
Engineering Journal of Science and Engineering, 24(72), 799-813.

Van Den Bergh, F. 2001. An Analysis of Particle Swarm Optimizers (PSO). Pretoria,
University of Pretoria, 78-85.

166
Vashisth, D., Srivastava, S. and Agarwal, A. (2018). Inversion of potential field data
using whale optimization. SEG Technical Program Expanded Abstracts, October
17.

Vasuki, A. 2020. Nature-Inspired Optimization Algorithms. CRC Press, Taylor &


Francis Group, Boca Raton, USA.

Walton S., Hassan O., Morgan K. and Brown M.R. (2011). Modified cuckoo search:
a new gradient free optimization algorithm. Chaos, Solitons & Fractals 44(9),
710-718.

Wang, G., Xu, C. and Liu, G. (2018). The transient electromagnetic inversion based
on the simplex-simulated annealing algorithm. Proceedings of the 37th Chinese
Control Conference, Wuhan, China, 4321-4324.

Wang, D., Tan, D. and Liu, L. 2018. Particle swarm optimization algorithm: an
overview. Soft Computing, 22(2), 387-408.

Wang, H., Liu, M.L., Xi, Z.Z., Peng, X.L. and He, H. (2018). Magnetotelluric
inversion based on BP neural network optimized by genetic algorithm. Chinese
Journal of Geophysics, 61(4), 1564-1575.

Wang, W., Xie, B., Pan, K., Wang, J. and Feng, B. (2010). Electromagnetic imaging
of heterogeneous media based on the accelerated differential evolution algorithm.
Progress in Geophysics, 25(6), 2002-2008.

Wang, Y., Sun, C. and Wu, D. (2018). Rayleigh wave nonlinear inversion based on
the Shuffled frog-leaping algorithm. International Geophysical Conference, 24-
27 April, Beijing, China.

Weber, Z. (2000). Seismic traveltime tomography: a simulated annealing approach.


Physics of the Earth and Planetary Interiors, 119, 149-159.

167
Wu, P., Tan, H., Peng, M., Ma, H. and Wang, M. (2018). Joint Inversion of 1-D
Magnetotelluric and Surface-Wave Dispersion Data with an Improved Multi-
Objective Genetic Algorithm and Application to the Data of the Longmenshan
Fault Zone. Pure and Applied Geophysics, 175, 3591-3604.

Wushouaili, R., Liang, S., Wang, Y. and Li, F. (2018). Magnetotelluric inversion for
symmetrically anisotropic layered medium using genetic algorithm. Progress in
Geophysics, 33(6), 2477-2482.

Xing, Z., Mazzotti, A. (2019a). Two-grid full-waveform Rayleigh-wave inversion via


a genetic algorithm — Part 1: Method and synthetic examples. Geophysics, 84(5),
805-814.

Xing, Z., Mazzotti, A. (2019b). Two-grid full-waveform Rayleigh-wave inversion via


a genetic algorithm — Part 2: Application to two actual data sets. Geophyiıcs,
84(5), 815-825.

Xiong, J., Liu, C., Chen, Y. and Zhang, S. (2018). A non-linear geophysical inversion
algorithm for the MT data based on improved differential evolution. Engineering
Letters, 26(1), 1-11.

Yan, L., Shen, Q., Lu, H., Wang, H., Fu, X. and Chen, J. (2020). Inversion and
uncertainty assessment of ultra-deep azimuthal resistivity logging-while-drilling
measurements using particle swarm optimization. Journal of Applied Geophysics,
178, 104059.

Yan, X., Li, P., Tang, K., Gao, L. and Wang, L. (2020). Clonal selection based
intelligent parameter inversion algorithm for prestack seismic data. Information
Sciences, 517, 86-99.

Yang, X.S. (2010). Nature-Inspired Metaheuristic Algorithms, 2nd edn. Luniver Press,
UK.

168
Yang, X.S. (2011). Metaheuristic Optimization. Scholarpedia, 6(8), 11472.

Yang, X.S. (2014). Nature-Inspired Optimization Algorithms, First Edition. Elsevier,


Massachusetts, USA.

Yang X-S. and Deb S. (2009). Cuckoo search via Lévy flights. In: IEEE World
Congress on Nature and Biologically Inspired Computing (NaBIC); Coimbatore,
India, 210-214.

Yang, X.S. and Deb S. (2013). Multiobjective cuckoo search for design optimization.
Computers & Operations Research. 40(6), 1616-1624.

Yang, X.S., Deb, S. and Fong, S. (2014). Metaheuristic Algorithms: Optimal Balance
of Intensification and Diversification. Applied Mathematics & Information
Sciences, 8(3), 977-983.

Yin, C. and Hodges, G. (2017). Simulated annealing for airborne EM inversion.


Geophysics, 72(4), 189-195.

Yogi, I.B.S. and Widodo (2017). Time domain electromagnetic 1D inversion using
genetic algorithm and particle swarm optimization. AIP Conference Proceedings,
39(3), 665-676.

Yue, B.B., Peng, Z.M., Hong, Y.G. and Wen, Z. (2009). Wavelet inversion of pre-
stack seismic angle-gather based on particle swarm optimization. Chinese Journal
of Geophysics, 52(12), 3116-3123.

Yu, P., Wang, J.L., Wu, J.S. and Wang, D.W. (2007). Constrained joint inversion of
gravity and seismic data using the simulated annealing algorithm. Chinese Journal
of Geophysics, 50(2), 465-475.

169
Yüngül S. (1950). Interpretation of spontaneous polarization anomalies caused by
spheroidal orebodies. Geophysics, 15(2), 237-246.

Zhdanov, M.S. (2015). Inverse Theory and Applications in Geophysics, Second


Edition. Elsevier, Amsterdam, Netherlands.

Zeng, Z.W., Chen, X., Yang, H.Y., Zhang, Z.Y., Guo, Y.H., Liu, X. and Ye, Y.X.
(2020). Joint inversion of magnetotelluric and gravity based on improved
differential evolution algorithm and wide range petrophysical constraints. Chinese
Journal of Geophysics, 63(12), 4565-4577.

Zhang, Y., Xu, Y. and Yang, B. (2021). Lévy gradient descent: augmented random
search for geophysical inverse problems. Surveys in Geophysics, 42, 899-921.

Zhu, L., Liu, E. and McClellan, J.H. (2015). Full waveform microseismic inversion
using differential evolution algorithm. IEEE Global Conference on Signal and
Information Processing.

Zuo, W. 1995. Multivariable adaptive control for a space station using genetic
algorithms. IEE Proceedings-Control Theory and Applications, 142(2), 81-87.

170
ProQuest Number: 30822576

INFORMATION TO ALL USERS


The quality and completeness of this reproduction is dependent on the quality
and completeness of the copy made available to ProQuest.

Distributed by ProQuest LLC ( 2023 ).


Copyright of the Dissertation is held by the Author unless otherwise noted.

This work may be used in accordance with the terms of the Creative Commons license
or other rights statement, as indicated in the copyright statement or in the metadata
associated with this work. Unless otherwise specified in the copyright statement
or the metadata, all rights are reserved by the copyright holder.

This work is protected against unauthorized copying under Title 17,


United States Code and other applicable copyright laws.

Microform Edition where available © ProQuest LLC. No reproduction or digitization


of the Microform Edition is authorized without permission of ProQuest LLC.

ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346 USA

You might also like