1D Inversion of vertical electrical sounding (VES) using a Whale Optimization

Algorithm (WOA).
Jorge L. Abril1, Cassiano A. Bortolozo2, Francisco Marcio Barboza3
1
CPGG/UFBA, 2CEMADEN, 3CERES
Copyright 2021, SBGf - Sociedade Brasileira de Geofísica physics-based and swarm-based methods. The most
This paper was prepared for presentation during the 17th International Congress of the popular evolution-inspired technique is Genetic
Brazilian Geophysical Society held in Rio de Janeiro, Brazil, 16-19 August 2021.
Algorithms (GA) (e.g., Goldberg, 1989). Physics-based
Contents of this paper were reviewed by the Technical Committee of the 17 th methods imitate the physical rules in the universe. The
International Congress of the Brazilian Geophysical Society and do not necessarily
represent any position of the SBGf, its officers or members. Electronic reproduction or most popular algorithms are Simulated Annealing (SA),
storage of any part of this paper for commercial purposes without the written consent of (e.g., Kirkpatrick, 1983).
the Brazilian Geophysical Society is prohibited.

_____________________________________________________________________ The third group of nature-inspired methods includes

Abstract swarm- based techniques that mimic the social behavior
of groups of animals. The most popular algorithm is
Particle Swarm Optimization, originally developed by
Direct current resistivity surveys have been widely used in (Kennedy and Eberhart 1995). This study describes a
mining, geotechnical and hydrogeological studies due to new meta-heuristic optimization algorithm (namely, Whale
the low cost and accuracy in mapping and detecting Optimization Algorithm, WOA) mimicking the hunting
structures in subsurface. In the last two decades, global behavior of humpback whales, described by (Mirjalili,
optimization methods are becoming more popular in 2016). This algorithm describes a technique of spiral to
geophysics because they do not require gradient simulate bubble-net attacking mechanism of humpback
information and no initial model to obtain a good result to whales. The efficiency of the WOA algorithm developed in
infer characteristics of the subsurface resistivity this expanded abstract is tested by solving two synthetic
distribution. In this research, we implemented a Whale models and a set of field data. Optimization results
Optimization Algorithm (WOA) that mimics the social demonstrate that WOA is very competitive compared to
behavior of the humpback whales. To demonstrate the the state of the art optimization methods.
robustness of the WOA, we test under three synthetic
models with 2% of Gaussian noise. The results shows an
efficient inversion process and a low uncertainty of
inverted estimated model. 1D forward modelling resistivity

To solve the 1D forward modelling of resistivity we adopt

Introduction the linear filtering method developed by (Ghost 1971).
This method solves the resistivity function integral by a
A wide quantity of optimization problems in geophysics convolution. The corresponding apparent resistivity
are not linear, multidimensional and characterized by ρa(d,m) is calculated by Koefoed, (1970), as follows in the
complex objective functions with several local minimum. equation (1) :
There are two techniques for solving these inverse
problems: deterministic methods based on gradient of the
objective function and stochastic methods, which perform
a random direct search in the solution space. These last, ∞ (1)
show a high demand of computational cost when the
forward modelling is expensive, also they can be trapped
ρa ( d , m )=ρ1 +d
2
∫ ( H 1 ( λ ,m )− ρ1 ) . J 1 (λd ) λdλ
0
in false global minimum, principally, in cases where it
presents a huge quantity of parameters in the model. Where d is the distance between current electrodes, J 1 is
Fortunately, 1D forward modelling of resistivity is cheap the first Bessel function, H1 is the resistivity transform
computationally and is a good candidate to apply this kind kernel for the first layer and λ is the integration variable.
of techniques. The resistivity kernel is calculated by the recurrence
formula from bottom to the surface, following the equation
Different studies reported in the literature to solve (2) and equation (3) :
problems using a linear approach (e.g., Menke 1984;
Tarantola 1987). These simple techniques require an
initial solution close to the true model for obtaining a good
H n (λ)=ρn (2)

convergence. Other techniques used by solving non-

linear optimization approaches have been testes (e.g.,
Sen et al., 1993; Chunduru et al., 1995). These non-linear H i ( λ )=(H i+ 1+ ρi . th(λ hi ))/(+ H i+1 .th (λ(3)hi)/ ρi )
approaches can find a satisfactory solution in a large
search space, but they consume a lot of time. Within where Hi(λ) is the transform resistivity for the ith layer,
these global optimization methods (GOM), can be th(λhi) is the hyperbolic tangent function. This recursive
selected in three main categories: evolution based

Seventeenth International Congress of the Brazilian Geophysical Society

 Abril, Bortolozo & Barboza 2
______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

formula goes upward until H1 is determinated for the d⃗ =|⃗

C . ⃗x p ( i )−⃗x hw (i)| (6)
surface layer.

Method A⃗
⃗x hw ( i+1 )=⃗x p ( i )−⃗ .d (7)

The inverse theory is the process to obtain from a set of

Where i is the ith iteration. The coefficient vectors are
observations the model that produced the observations.
determinated as follows in the equation (8) and equation
In other words, the observations of field or collected data
(9), respectively:
give a set of data vector d= [d1, d2, d3…dn]T, which
corresponds to a m model vector, m= [m1, m2, m3…mn]T.
These vectors are related through the operator g, in the ⃗
A=2. ⃗a . r⃗1−⃗a (8)
following equation (4):

d=g (m) (4)

⃗
C =2. ⃗
r2 (9)

In the inversion process, we estimate a model mest that

minimizes the cost function, which is a measure of
discrepancy between the observed and computed data
sets. In this abstract, we use the normalized root square
Where components a ⃗ are lianarly decreased from 2 to 0
mean error (NRMS), as depicted in the equation (5) over the iterations, and r⃗ 1 and r⃗ 2 are random vectors in
the range [0,1].

√
(5)
1
N
d obs ( i )2−d cal (i)2 In this sense, the search of agents will move around the
NRMS= ∗∑ best solution obtained so far. The humpback whales use
N i d obs (i)2 the mechanism of bubble-net to attack the prey. This
behavior is formulated as follows:

Bubble-net attack mechanism:

Where dobs(i) and dcal(i) are observed and calculated data,
respectively, and N is the total number of data. For both This behavior is grouped in two principal strategies:
cases synthetic and real data we calculated the
normalized model NRMS misfit using the above equation. i) Shrinking encircling: This mechanism is
achieved by decreasing the value of a ⃗ in the
equation (8).This value define the strategy
Whale optimization Algorithm (WOA) for searching (⃗A >1) and attacking (⃗ A <1)
the prey.
Bases on the social behavior of the whales in nature,
ii) Spiral updating position: This behavior
hump whales dive around the prey and then start to
calculate the distance between the whale
create bubble in a spiral shaped and swim up toward the
and the prey as follows in the equation (10):
surface. This behavior is unique in nature and this
maneuver of humpback whales is defined mathematically
⃗x hw ( t +1 )=d⃗ . e . cos ( 2 πl ) + ⃗x p (t ) (10)
' bl
in order to perform optimization problem.

Mathematical expressions and optimization algorithm

Where ⃗ |
d ' = ⃗x p ( t )−⃗x h w (t) indicatres the position of ith |
Here, we presented the principal equations related to the whale to the prey, b is a constant defining the shape of
mathematical model used by hump whales for searching logarithmic spiral and l is a random number between [-1,
the prey and simulate the mechanism of optimization, 1].
following three principals rules of encircling the prey,
spiral bubble-net maneuver and search for the prey: To define the quantity of exploration and exploitation if the
(WOA) we define a probability of 50% to choose either
Encircling the prey: the shrinking encircling behavior or the spiral model to
update the position of whales during optimization. This
Following Mirjalili, (2016), encircling behavior is mathematical representation for p <50% is as follows in
mathematically represented by the distance (d), position equation (11):
vectors of the prey ( ⃗
x p) and humpback whale ( ⃗x hw), and
using the coefficient vectors and ⃗
A and⃗
C related to the A . d⃗
⃗x hw (t +1)=⃗x p− ⃗ (11)

whales and the prey as showed in the equation (6) and

equation (7): And the representation when p>50% is depicted in the
equation (12),

Seventeenth International Congress of the Brazilian Geophysical Society

 Abril, Bortolozo & Barboza 3
______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

⃗x hw (t +1)=⃗ (12)
bl
d ' . e . cos ( 2 πl )+ ⃗x p ( t )
For all synthetic examples we set up the constants b and l
as 1 and a random number between [-1,1], respectively.

Seventeenth International Congress of the Brazilian Geophysical Society

 Abril, Bortolozo & Barboza 4
______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

Exploration mechanism (Search for prey)

The mathematical model used for WOA to

emphasize exploration is represented as follows
in the equation (13):

d⃗ =|⃗ X rand −⃗x hw|

C .⃗ (13)

Where ⃗x hw is expressed by the equation (14)

⃗x hw (t +1)=⃗ A . ⃗d
X rand −⃗ (14)

Using the same mechanism based on updating

the position of whales according to the position
of each other, we update the position of the
search agent according to randomly chosen
search agent instead the best search agent
found so far. Figure 2 Misfit between observed and calculated data
using the best model obtained with WOA. History of
convergence using WOA.
Inversion of synthetic data

We considered layered earth models for testing The misfit between data and model responses
the effectiveness of WOA. We designed two are calculated through the relative error, given
types of models, (H-type model) and (K-type as follows in the equation (15):
model). The comparative study is summarized in

| |
Table 2 and Table 3. Schlumberger vertical N
d i− y i (15)
100
electrical soundings were generated adding misfit= ∗∑
random Gaussian noise of 2% to direct current N i=1 di
resistivity raw data. Initial guesses used for
delimiting the lower bound and upper bound of
the search space were made based on a
qualitative interpretation of the apparent All calculated models have misfit values less
resistivity curves. The Figure 1 and the Figure 2 than 1%. In the other hand, we build a synthetic
show a comparison between the true model and model corresponding to stratigraphy of
the best model obtained using (WOA) for the H- Bebedouro Region in Paraná Sedimentary
type model and K-type model, respectively. In Basin, in Brazil, following a real geoelectric
addition, these Figures include the history of stratigraphy observed in that area (Porsani, et
convergence using the WOA. al., 2012.). This stratigraphy is composed by 4
layers: The initial layer with 8 m of thickness and
200 Ω.m correspond a layer with unsaturated
soil. The second layer composed by a saturated
zone is characterized by 25 Ω.m and 55 m thick.
The resistivity layer below 800 Ω.m and 500 m
thick is represented by basaltic rock. The last
layer correspond to Guarani aquifer with 30 Ω.m,
as described in (Bortolozo, et al., 2015). Similar
to a two synthetic models presented above, we
define the search space based on apparent
resistivity curves Table 1 and the results of the
inversion process are showed in the Figure 3.
The misftit between data and model responses
have a misfit less than 0.08%.

Figure 1 Misfit between observed and calculated data

using the best model obtained with WOA. History of
convergence using WOA.

Seventeenth International Congress of the Brazilian Geophysical Society

 Abril, Bortolozo & Barboza 5
______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

We analyzed the performance of the Whale

optimization algorithm (WOA) applied in a set of
synthetic data created based on standard curves
of resistivity data and a set of synthetic
geoelectric data from Paraná Basin in Brazil. We
use 7 whales and 100 iterations to achieve a
good performance between the true model and
the real inverted model with WOA. The misfit of
data gives an excellent adjust and offers an
additional alternative to obtain among the
traditional global optimization methods widely
studied in the literature, corresponding to 1D
inversion of resistivity data.

Acknowledgments

We thankful to PROAP founding for providing

financial support to participate in this event.
Figure 3 Misfit between observed and calculated data
using the best model obtained with WOA. History of References
convergence using WOA.
BORTOLOZO, C. A., PORSANI, J. L., DOS
SANTOS, F. A. M., & ALMEIDA, E. R.
(2015). VES/TEM 1D joint inversion by
Parameters True Min. Max.
using Controlled Random Search (CRS)
Value
ρ1 (Ω.m) 100 5 200
algorithm. Journal of Applied
ρ2 (Ω.m) 10 1 30 Geophysics, 112, 157-174.
ρ3 (Ω.m) 500 100 1000
h1 (m) 5 0.1 10 CHUNDURU, R. K., SEN, M. K., STOFFA,
h2 (m) 15 1 30 P. L., & NAGENDRA, R. (1995). Non‐linear
Table 2True parameters and search range space for t inversion of resistivity profiling data for
he inversion of H-type synthetic data using WOA. some regular geometrical
bodies1. Geophysical Prospecting, 43(8),
Parameters True Min. Max. 979-1003.
Value
ρ1 (Ω.m) 20 5 120 GHOSH, D. P. (1971). The application of
ρ2 (Ω.m) 1000 50 3000
linear filter theory to the direct interpretation
ρ3 (Ω.m) 10 1 120
of geoelectrical resistivity sounding
h1 (m) 15 2 35
h2 (m) 5 0.5 20
measurements. Geophysical
Table 3. True parameters and search space range for prospecting, 19(2), 192-217.
the inversion of K-type synthetic data using WOA.
GOLDBERG, D. E., & SASTRY, K. (1989,
January). A practical schema theorem for
Parameters True Min. Max. genetic algorithm design and tuning.
Value In Proceedings of the genetic and
ρ1 (Ω.m) 200 100 300 evolutionary computation conference (Vol.
ρ2 (Ω.m) 25 10 50 328335).
ρ3 (Ω.m) 800 500 1000
ρ4 (Ω.m) 30 10 50 KENNEDY, J., & EBERHART, R. (1995,
h1 (m) 8 1 10 November). Particle swarm optimization.
h2 (m) 55 30 60 In Proceedings of ICNN'95-international
h3 (m) 500 100 800 conference on neural networks (Vol. 4, pp.
Table 3. True parameters and search space range for 1942-1948). IEEE.
the inversion of Bebedouro Basin using WOA.
KIRKPATRICK, S., GELATT, C. D., &
VECCHI, M. P. (1983). Optimization by
simulated annealing. science, 220(4598),
671-680.

KOEFOED, O. (1970). A fast method for

Conclusions determining the layer distribution from the

Seventeenth International Congress of the Brazilian Geophysical Society

 Abril, Bortolozo & Barboza 6
______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

raised kernel function in geoelegtrical

sounding. Geophysical Prospecting, 18(4),
564-570.

MENKE, W. (1984). Geophysical data

analysis: Discrete inverse theory. Academic
press.

MIRJALILI, S; LEWIS, A. (2016). The

Whale optimization algorithm. Advances in
engineering software, 95, 51-67.

PORSANI, J. L., ALMEIDA, E. R.,

BORTOLOZO, C. A., & DOS SANTOS, F.
A. M. (2012). TDEM survey in an area of
seismicity induced by water wells in Paraná
sedimentary basin, Northern São Paulo
State, Brazil. Journal of applied
geophysics, 82, 75-83.

SEN, M. K., & STOFFA, P. L. (1991).

Nonlinear one-dimensional seismic
waveform inversion using simulated
annealing. Geophysics, 56(10), 1624-1638.

SEN, M. K., BHATTACHARYA, B. B., &

STOFFA, P. L. (1993). Nonlinear inversion
of resistivity sounding
data. Geophysics, 58(4), 496-507.

TARANTOLA, A. (1987). Inversion of travel

times and seismic waveforms. Seismic
tomography, 135-157.

Seventeenth International Congress of the Brazilian Geophysical Society