Bajece Vol-4 No-2 Full

BAJECE
emai
l
:edi
t
or
@baj
ec
e
.
c
om
Bal
kanJournalof
El
ect
r
i
c
al
&Comput
er
Engi
neer
i
ng
www.
ba
j
e
c
e
.
c
om
AnI
nt
er
nat
i
onalPeerRevi
ewed,Ref
er
r
ed,I
ndexedandOpenAcces
sJ
our
nal
Vol
:
4
No:
2
Year:
2016
I
SSN:
2147-284X
Sponsored by the
Istanbul Technical University
This journal is accredited by the Istanbul Technical
University subsidy purposes. It is abstracted and
indexed in, Index Google Scholarship, the PSCR,
DOAJ, Research Bible, Indian Open Access Journals (OAJ),
Institutional Repositories (IR), J-Gate (Informatics India), Ulrichs,
ResearchGate, International Society of Universal Research in Sciences,
DRJI, EyeSource.
General Publication Director & Editor-in-Chief
.Serhat Seker, Istanbul Technical University, Turkey
Vice Editor
Amir Toki, University of Tuzla, Bosnia and Herzegovina
Tahir Cetin Akinci, Istanbul Technical University, Turkey
Veselina Nedeva, Trakia University, Bulgaria
BAJECE
ISSN: 2147- 284X

Vol: 4
No : 2
Year: September 2016
Scientific Committee
Abhishek Shukla (India)
Aleksandar Georgiev (Bulgaria)
Arunas Lipnickas (Lithuania)
Audrius Senulis (Lithuania)
Belle R. Upadhyaya (USA)
Brijender Kahanwal (India)
Daniela Dzhonova-Atanasova (Bulgaria)
Deris Stiawan (Indonesia)
Emine Ayaz (Turkey)
Enver Hatimi (Kosovo)
Ferhat Sahin (USA)
Gursel Alici (Australia)
Hakan Temelta (Turkey)
Ibrahim Akduman (Turkey)
Jan Izykowski (Poland)
Javier Bilbao Landatxe (Spain)
Jelena Dikun (Lithuania)
Kunihiko Nabeshima (Japan)
Lambros Ekonomou (Greece)
Marcel Istrate (Romania)
Marija Eidukeviciute (Lithuania)
Milena Lazarova (Bulgaria)
Muhammad Hadi (Australia)
Muhamed Turkanovi (Slovenia)
Mourad Houabes (Algerie)
Murari Mohan Saha (Sweden)
Okyay Kaynak (Turkey)
Osman Nuri Ucan (Turkey)
Ozgur E. Mustecaplioglu (Turkey)
Padmanaban Sanjeevikumar (India)
Ramazan Caglar (Turkey)
Rumen Popov (Bulgaria)
Tarek Bouktir (Algeria)
Sead Berberovic (Croatia)
Seta Bogosyan (USA)
Savvas G. Vassiliadis (Greece)
Suwarno (Indonesia)
Tulay Adali (USA)
Vitalijus Volkovas (Lithuania)
Yogeshwarsing Calleecharan (Mauritius)
YangQuan Chen (USA)
Youcef Soufi (Algeria)
CONTENTS
R.Y. Kazakova, S.I. Nedelcheva, and R.K. Popov;

Estimation of Wind Turbine Generator Model
Parameters using Artificial Intelligence Methods, . 51-57
D. Turkpence; Simulation of Entanglement Dynamics
of Flying Qubits Through Photonic Fields, 58-61
K.Dawood; Hybrid wind-solar reliable solution for
Turkey to meet electric demand, .....62-66
S. Singh and S. Bhullar; Hardware implementation of
auto switching and light intensity control of LED lamps,
................................................................................ 67-71
D.K. Nedev, and K.D. Nedev; Aeronautical
telecommunication simulator, .72-74
S. Hansun; A New Approach of Browns Double
Exponential Smoothing Method in Time Series
Analysis.. 75-78
C. Bakir; Automatic Speaker Gender Identification for
the German Language, ... 79-83
G. Kavuran, B. B. Alagoz, A. Ates, C. Yeroglu;
Implementation of Model Reference Adaptive Controller
with Fractional Order Adjustment Rules for Coaxial
Rotor Control Test System, .84-88
H.S. Nogay; Determination of Leakage Reactance in
Monophase Transformers Using by Cascaded Neural
Network, .................................................................89-96
Aim & Scope

The journal publishes original papers in the extensive field of
Electrical-Electronics and Computer engineering. It accepts
contributions which are fundamental for the development of electrical
engineering, computer engineering and its applications, including
overlaps to physics. Manuscripts on both theoretical and experimental
work are welcome. Review articles and letters to the editors are also
included.
Application areas include (but are not limited to): Electrical &
Electronics Engineering, Computer Engineering, Software
Engineering, Biomedical Engineering, Electrical Power Engineering,
Control Engineering, Signal and Image Processing, Communications &
Networking, Sensors, Actuators, Remote Sensing, Consumer
Electronics, Fiber-Optics, Radar and Sonar Systems, Artificial
Intelligence and its applications, Expert Systems, Medical Imaging,
Biomedical Analysis and its applications, Computer Vision, Pattern
Recognition, Robotics, Industrial Automation.
BALKAN
JOURNAL OF
ELECTRICAL & COMPUTER
ENGINEERING
(An International Peer Reviewed, Indexed and Open Access Journal)
Contact

Department of Electrical Engineering
Ayazaga Campus, Maslak, Istanbul-Turkey
Web: https://www.bajece.com
http://dergipark.ulakbim.gov.tr/bajece/
e-mail: editor@bajece.com
BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING
DOI: 10.17694/bajece.52662
51
Estimation of Wind Turbine Generator Model

Parameters using Artificial Intelligence Methods
R.Y. Kazakova, S.I. Nedelcheva, R.K. Popov
Abstract Modelling (in a broad sense) is an essential tool for

research in all areas and represents a scientifically based method
for assessing the performance of systems and processes used for
making engineering decisions. This applies in particular to the field
of management systems, where the foundation is making decisions
based on the information received.
The existing and newly designed systems effectively examined by
using the mathematical models (analytical and spoofing) which
allows identifying some constant parameters that are involved in
the differential equations representing the dynamics of the system
analyzed. Such systems may come from a broad scientific spectrum,
for example from economics and biology from communication and
weather forecasting.
The present paper investigates some Artificial Intelligence (AI)
methods identifying the parameters of a dynamical system. Two
types of methods are compared - 'evolution' and 'particle swarm'
intelligence. First, for this purpose, a system simulation model
generating data (for the two methods of identification in order to
compare afterwards the results) is used. After that, Genetic (GA)
and Particle Swarm Optimization (PSO) algorithms are applied to
estimate the wind turbine generator model parameters. The results
of both methods are compared in terms of their accuracy and
performance. The software for the simulation and AI process has
been developed using MATLAB.
Index Terms Artificial intelligence, system parameter
estimation, genetic algorithm, particle swarm optimization, wind
turbine generator system model
I.
INTRODUCTION
main focus in system identification is on the parameter

THE
estimation process. Well-developed techniques such as
least-square (LS), instrumental variable and maximum
likelihood exist for parameters estimation of models.
However, these techniques often fail in search for the global
optimum if the search space is not differentiable or linear in
the parameters.
To date, artificial intelligence (AI) techniques have become
potential candidates to many control applications. One of the
R. KAZAKOVA is with the Department of Optoelectronics and Laser
Engineering, Technical University of Sofia, Branch Plovdiv, Bulgaria (email: rrrrosi@abv.bg).
S.I NEDELCHEVA is with the Department of Electrical Engineering,
Electronics and Automation in Faculty of Engineering and PedagogySliven at Technical University of Sofia. (e-mail: stefned@abv.bg).
R. POPOV is with the Department of Electronics, Communications and
Information Technologies, Plovdiv University Paisii Hilendarski, Bulgaria
(e-mail: rum_pop@yahoo.com).
Copyright BAJECE
ISSN: 2147-284X
most powerful AI techniques is genetic algorithm (GA),

which has been widely used and applied to control systems
[1]. Genetic algorithms are very good for optimization
problems with several local minima where conventional
search algorithms fail. GA techniques can be effectively
applied to system identification problem to estimate the
model parameters. A GA simultaneously evaluates many
points in the parameters space and converges toward the
global solution. It does not require the search space to be
differentiable or continuous [3,4]. Many researchers have
applied the GA techniques to identify linear and non-linear
systems. The wind turbine generator system is a complex
nonlinear system with parameters that are difficult to identify
using standard LS techniques.
II. ESTIMATION METHODS
A. Genetic algorithm overview
The genetic algorithm is a method for solving both
constrained and unconstrained optimization problems that is
based on natural selection, the process that drives biological
evolution. The genetic algorithm repeatedly modifies a
population of individual solutions. At each step, the genetic
algorithm selects individuals at random from the current
population to be parents and uses them to produce the
children for the next generation. Over successive generations,
the population "evolves" toward an optimal solution. We can
apply the genetic algorithm to solve a variety of optimization
problems that are not well suited for standard optimization
algorithms, including problems in which the objective
function is discontinuous, non-differentiable, stochastic, or
highly nonlinear.
The genetic algorithm uses three main types of rules at
each step to create the next generation from the current
population:
Selection rules select the individuals, called parents,
which contribute to the population at the next
generation.
Crossover rules combine two parents to form
children for the next generation.
Mutation rules apply random changes to individual
parents to form children.
GA was pioneered in 1975 by Holland [5], and its concept
is to mimic the natural evolution of a population by allowing
solutions to reproduce, creating new solutions, which then
compete for survival in the next iteration. The fitness
September 2016
Vol.4 No.2
http://www.bajece.com
improves over generation sand the best solution is finally

achieved. The initial population, P(0), is encoded randomly
by strings. In each generation, t, the more fit elements are
selected for the mating pool; and then processed by three
basic genetic operators, reproduction, crossover, and
mutation, to generate new offspring. On the basis of the
principle of survival of the fittest, the best chromosome of a
candidate solution is obtained. The pseudo code of GA
illustrates the procedure of the computation as follows:
Procedure GA
begin
t=0
initialize P(t)
evaluate P(t)
while not satisfy stopping rule do
begin
t= t+1
select P(t) from P(t1)
alter P(t)
evaluate P(t)
end
end
The power of GA lies in its simultaneous searching a

population of points in parallel, not a single point. Therefore
GA can find the approximate optimum quickly without
falling in to a local optimum. In addition GA does not have
the limitation of differentiability, as do the mathematical
techniques. These characteristics of GA are the reasons it is
used here for the problem of model identification in ARIMA
models.
B. Concepts of the PSO approach
The neural network is a simplified model of human brain;
genetic algorithm is inspired by the human evolution. Here
we discuss another type of biological system - social system,
more specifically, the collective behaviors of simple
individuals interacting with their environment and each other.
Someone called it as swarm intelligence. All of the
simulations utilized local processes, such as those modeled
by cellular automata, and might underlie the unpredictable
group dynamics of social behavior.
There are two popular swarm inspired methods in
computational intelligence areas: Ant colony optimization
(ACO) and particle swarm optimization (PSO). ACO was
inspired by the behaviors of ants and has many successful
applications in discrete optimization problems [6].
The particle swarm concept originated as a simulation of
simplified social system. The original intent was to
graphically simulate the choreography of bird of a bird block
or fish school. However, it was found that particle swarm
model can be used as an optimizer [7].
Particle swarm optimization (PSO) is a population based
stochastic optimization technique developed by Dr. Eberhart
[8] and Dr. Kennedy in 1995, inspired by social behavior of
bird flocking or fish schooling.
PSO shares many similarities with evolutionary
computation techniques such as GA. The system is initialized
with a population of random solutions and searches for
optima by updating generations. However, unlike GA, PSO
has no evolution operators such as crossover and mutation. In
Copyright BAJECE
ISSN: 2147-284X
DOI: 10.17694/bajece.52662
52
PSO, the potential solutions, called particles, fly through the

problem space by following the current optimum particles.
As stated before, PSO simulates the behaviors of bird
flocking. Suppose the following scenario: a group of birds are
randomly searching food in an area. There is only one piece
of food in the area being searched. All the birds do not know
where the food is. But they know how far the food is in each
iteration. So what's the best strategy to find the food? The
effective one is to follow the bird which is nearest to the food.
PSO learned from the scenario and used it to solve the
optimization problems. In PSO, each single solution is a
"bird" in the search space. It is called "particle". All of
particles have fitness values which are evaluated by the
fitness function to be optimized, and have velocities which
direct the flying of the particles. The particles fly through the
problem space by following the current optimum particles.
PSO is initialized with a group of random particles
(solutions) and then searches for optima by updating
generations. In every iteration, each particle is updated by
following two "best" values. The first one is the best solution
(fitness) it has achieved so far. (The fitness value is also
stored.) This value is called pbest. Another "best" value that is
tracked by the particle swarm optimizer is the best value,
obtained so far by any particle in the population. This best
value is a global best and called gbest. When a particle takes
part of the population as its topological neighbors, the best
value is a local best and is called lbest.
After finding the two best values, the particle updates its
velocity and positions with following equations (1) and (2).
v[ ] = v[ ] + c1 * rand( )*(p best[ ] - present[ ]) +
+c2*rand()*(gbest[ ] - present[ ]) (1)
present[ ] = persent[ ] + v[ ]
(2)
v[ ] is the particle velocity, persent[ ] is the current particle

(solution). pbest[ ] and gbest[ ] are defined as stated before.
Rand ( ) is a random number between (0,1). c1, c2 are
learning factors. Usually c1 = c2 = 2. The pseudo code of the
procedure is as follows:
for each particle
Initialize particle
end
do
for each particle
Calculate fitness value
if the fitness value is better than the best fitness value
(pbest) in history
set current value as the new pbest
end
Choose the particle with the best fitness value of all the
particles as the gbest
for each particle
Calculate particle velocity according equation (1)
Update particle position according Equ. (2)
end
while maximum iterations or minimum error criteria is not

attained
Particles' velocities on each dimension are clamped to a
maximum velocity Vmax. If the sum of accelerations would
cause the velocity on that dimension to exceed Vmax, which is
a parameter specified by the user, then the velocity on that
dimension is limited to Vmax.
September 2016
Vol.4 No.2
III. MODEL DESCRIPTION

The system, modeled in this article represents a wind
turbine generator. It consists of three main subsystems: wind
turbine; synchronous generator and mechanical part.
A. Mathematical description
The mechanical subsystem is represented by the Simulinkblocks which correspond to the equation:
=
(3)
For a description of processes in the synchronous generator

using the model of generalized electric machine axes d and q
(d axis coincides with the axis of the poles and q-axis with
geometric neutral line). The axes d and q are considered
connected with the poles of the rotor. Windings of the
armature are replaced with two equivalent in effect windings
the axes d and q with respective electromotive forces -
DOI: 10.17694/bajece.52662
53
voltage, resulting from the relative rotation of the actual stator

windings in the coordinate system of the rotor.
The mathematical description of the dynamic processes in
the synchronous electric drive is obtained by the equations
write mechanical properties at axes d and q, associated with
the rotor, on which is disposed exciting coil:
(4)
where:
(5)
is equal to zero.
The output power of the aerodynamic the wind turbine is
described by the following equation [2]:
P = (, )
3
2
(6)
where:
Pm
cp
- Mechanical output power of the turbine (W);

- Performance coefficient of the turbine;
- Air density (kg/m3);
- Turbine swept area (m2);
- Wind speed (m/s);
- Tip speed ratio of the rotor blade tip speed to wind speed
- Blade pitch angle (deg).
Turbine Power Characteristics (Pitch angle beta = 4 deg)

Turbine output power (pu of nominal mechanical power)
is the modulus of firmness of the asynchronous

component to a moment determined by the operation of the snubber
winding. In our case the snubber winding absent and the component
1.2
14.4 m/s
0.8 Max. power at base wind speed (12 m/s) and beta = 0 deg
0.6
13.2 m/s
12 m/s
10.8 m/s
0.4
9.6 m/s
0.2
8.4 m/s
7.2 m/s
6 m/s
0
0
0.2
1 pu
0.4
0.6
0.8
1
1.2
Turbine speed (pu of nominal generator speed)
1.4
Fig. 1. Power characteristics
Equation (6) can be normalized. In the per unit (pu) system

we have:
,
(7)
where :
Pm_pu - Power in pu of nominal power for particular values of and
A;
cp_pu - Performance coefficient in pu of the maximum value of cp;
wind_pu - Wind speed in pu of the base wind speed. The base wind
speed is the mean value of the expected wind speed in m/s;
kp - Power gain for cp_pu = 1 and wind_pu = 1 pu, kp is less than or equal
to 1
B. The model structure

The structural scheme of the mathematical model of the
wind generator is shown in Fig.2. The system consists of a
generator, mechanical subsystem and a wind turbine.
The mechanical subsystem is represented by Simulinkblocks Sum, Gain2 and Int1, which correspond to equation (3).
The scheme of the model of the synchronous generator is
shown in Fig.3. It is made up on the basis of the system of
equations (4) and (5).
A generic equation is used to model cp (,). This equation,

based on the modeling turbine characteristics of [1], is:
(8)
with
(9)
The coefficients c1 to c6 are: c1 = 0.5176, c2 = 116, c3 = 0.4,
c4 = 5, c5 = 21 and c6 = 0.0068.
The mechanical power Pm as a function of generator speed,
for different wind speeds and for blade pitch angle = 4
degree and base wind speed = 12 m/s , is illustrated below
Fig. 1.
Copyright BAJECE
ISSN: 2147-284X
Fig. 2. The structural scheme of the mathematical model of the wind

generator
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.52662
54
model to be identified: Rsum , Rf , L12, L1, J . Monte Carlo

simulation planning using central composite design is
performed. Bounds for each parameter are defined in the
parameter space. Two algorithms: real coded GA and PSO are
applied and tested. Firstly, we have to choice input factors to
be varied. In GA algorithm they are:
Crossover_p Crossover probability parameter;

Mutarion_r Mutation rate.
Varried parameters in PSO algorithm are chosen to be:
Fig. 3. The structural scheme of the synchronous generator mathematical

model
For a description of the wind turbine a library model of

Matlab [Simulink-Sim Power Systems], [2] has been used
IV. SIMULATIONS
The basic simulation was performed at the following
parameter values:
Estimated parameters:
Rsum = 0.894 - Sum of armature and load resistance (Ohm);
Rf = 2.524; - Excitation armature resistance (Ohm);
L12 = 0.152 - Mutual inductance (H);
L1 = 0.0117 - Own synchronous inductance (H);
J = 0.02
- Total inertia (kg.m2)
Other input parameters:
Correction_factor- is the correction (learning)

factor;
Inertia- is the inertia factor.
In the next step central composite design is performed. Only

one central point is used in case to obtain uniform values
distribution in the parameter space. For this purpose, is used
Matlab function ccdesign. Design results are showed on Fig.
5. Function plotGaSim.m is used here to display results.
After that parameter variance bounds are chosen:
Crossover_p = 0 1;
Mutation_r = 0.05 0.5
Generator excitation voltage = 50 V;

Initial rotor speed 100 rad/s;
Wind Speed = 12 m/s;
Pitch Angle = 4 deg;
The simulation results are illustrated in Fig.4

Fig. 5. Two factors central composite design test matrix
400
Id
Iq
300
The position of the factors in the space of the parameters in GA is

shown in Fig.6.
Current, A
200
100
-100
-200
-300
0.1
0.2
0.3
0.4
0.5
Time, s
0.6
0.7
0.8
0.9
Fig. 4. Simulation of change of currents along the axes d and q
Experiment Design
System, with parameters have to be estimated is represented
by the wind turbine generator system model, produced in
Simulink environment as a block diagram. This model was
described in details above. There are five parameters in system
Copyright BAJECE
ISSN: 2147-284X
Fig.6. Positions of the input variables in the parameter space in GA tests.
September 2016
Vol.4 No.2
Central composite design for the PSO test parameters variation

in Monte Carlo simulation are presented on fig 7.
DOI: 10.17694/bajece.52662
55
configuration
of
the
evaluated
The syntax of the function is as follows:
parameters.
fit = fitness5(gene),
where the gene is a vector containing the current set of genes (model
parameters).
V. RESULTS
Test results for first 5 estimations, using real coded GA and PSO
are reported in Table 1 and Table 2 respectively.
crossover_p
is the crossover parameter
mutation_r
is the mutation rate;
n_generations
is the number of generations.
The two parameters in PSO-based search algorithm may vary too:
correction factor and inertia. The related code is written as a Matlab
function with the next syntax:
Each algorithm was evaluated 90 times with varied parameters.

Evaluation function setting of the model is the same for GA and
PSO. It is based on a comparison of the outputs of the model and
evaluation model (changes in the values of the currents Id and Iq in
time). First configure the evaluation model on which to set the value
of the current set of genes (model parameters). Then he started end
to give the current realization of output functions. Finally, calculate
the average of the sum of absolute error AvSAE the current
Copyright BAJECE
ISSN: 2147-284X
Rsum,
Rf,
L12,
L1,
Ohm
Ohm
Kg.m2
0.83969
0.98713
0.53758
0.69762
0.48262
2.1095
3.0567
0.9729
1.4955
0.6502
0.14288
0.31685
0.074
0.07375
0.05
0.01199 0.0812
0.0236 0.01458
0.01351
0.00768 0.00523
0.00609 0.0095
TABLE II
MONTE CARLO SYSTEM PARAMETER ESTIMATION TEST RESULTS USING PSO
1
2
3
4
5
5.493
70.32
4.102
7.983
15.566
2
1.6
2
2.4
2.4
0.8
0.6
0.8
0.6
0.8
Rsum,
Rf,
L12,
L1,
Ohm
Ohm
Kg.m2
1.0016
3.1422
0.9912
1.0482
0.9133
2.5279
19.9406
2.5451
2.5577
2.4368
0.61985
0.99357
0.40782
0.8053
0.54215
0.0481
0.01348
0.0316
0.06334
0.04386
0.01592
0.2
0.01449
0.005
0.15934
VI. DISCUSSION
The significance analyzes for the varied parameters was
preformed too. Combined scatter plot for GA and PSO
parameters significance analyzes is showed on Fig.8 and Fig.9
respectively. Best values for the crossover parameter and
mutation rate are 0,6 and 0,5. Best values for the correction
factor and inertia are 2,0 and 0,8.
[A B AvSAE]=pso_proc(swarm_size, correction_factor,
inertia, iterations, err), where:
Scatter plot AvSAE vs. Crossover parameter and Mutation rate
100
80
AvSAE
A, B, AvSAE and err are the same as in function ga_proc;

swarm_size
is the size of the swarm;
correction_factor is the correction (learning) factor;
inertia
is the inertia factor
iterations
is the number of iterations.
AvSAE final
value
0.05 8.7812
0.5 25.5266
0.275 46.705
0.5 37.1226
0.05 57.3622
AvSAE final
value
A
is a vector, containing evaluated 'a' parameters;
B
is a vector, containing evaluated 'b' parameters (for
future use);
AvSAE is a vector of average sum of absolute error of the
model for all generations;
err
is an error tolerance;
popsize
is the size of population;
Mutation rate
The real coded GA search algorithm uses BLX- crossover

operation and two varied parameters: crossover parameter and
mutation rate. It is written as a Matlab function with the next
syntax:
[A B AvSAE]=ga_proc(err, popsize, crossover_p,
mutation_r, n_generations), where:
0.6
0.2
1
1
1
Run Number
Correction
factor
I nertia
Fig.7. Positions of the input variables in the parameter space in PSO tests
1
2
3
4
5
Crossover
probability
Run Number
TABLE I
MONTE CARLO SYSTEM PARAMETER ESTIMATION TEST RESULTS USING GA
60
40
20
0
0.2
0.4
0.6
0.8
1
Crossover parameter
0.1
0
0.2
0.3
0.4
0.5
0.6
Mutation rate
Fig. 8 Scatter plot of AvSAE vs. Crossover parameter and Mutation rate in
GA tests.
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.52662
56
Sorted AvSAE in GA and PSO

110
PSO
GA
100
Scatter plot AvSAE vs. Correction factor and Inertia
90
80
70
AvSAE
100
AvSAE
80
60
50
60
40
40
30
20
20
0
1.4
10
1.6
1.8
2
2.2
2.4
Correction factor
0.4
0.3
0.5
0.6
0.7
0.8
0.9
Provided null hypothesis F-test for each of tested

algorithms returns the result as listed below:
GA: F > P for all varied factors
Crossover parameter is significant: F = 7.5913, P = 0,1764.
Mutation rate is significant: F = 13.5861, P = 0,1764
PSO: F > P for all varied factors
Correction Factor is significant: F = 1.4571, P = 0,1764
Inertia is significant: F = 0.1955, P = 0,1764
The average values of the parameters calculated by GA and
PSO algorithm are showed in the following Table 3. It may be
noted that in this case PSO gives more accurate results. Are
taken over the values of the rows of tables with minimum error
AvSAE (Average Value of Sum of Absolute Error) respectively 79 rows of PSO Table 2 and row 27 of GA of
Table 1.
TABLE III
MEAN VALUES OF THE ESTIMATED PARAMETERS BY GA AND
PSO ALGORITHMS
Rsum
0.894
0.9362
0.88925
- 4.72
- 0.53
Rf
2.524
2.4347
2.5581
- 3.53
- 1.35
L12
0.152
0.18024
0.15901
-18.57
- 4.61
L1
0.0117
0.01439
0.01212
- 22,99- 3.58
J
0.02
0.02089
0.0195
- 4.45
2.5
The next Fig.10 is showed sorted values of AvSAE for both

algorithms PSO (in red) and GA (in blue).
Copyright BAJECE
10
20
30
40
50
60
Sorted Run number
70
80
90
Fig.10. Sorted by values of AvSAE for both algorithms PSO (in red) and
GA (in blue)
Inertia
Fig.9. Scatter plot of AvSAE versus Correction factor and Inertia in PSO
tests
Original
PSO
GA
Error PSO, %
Error GA, %
ISSN: 2147-284X
Test duration for all 90 estimations in each case (GA and

PSO) was measured to check algorithms' performance and the
reported results are:
GA test duration 2 h, 32 min, 50.7 s
PSO test duration 2 h, 37 min, 18.6 s
VII. CONCLUSIONS
1. Both GA and PSO algorithm are suitable for use in the
evaluation process the parameters of the wind turbine
generator system.
2. Table 3 and Fig.10 showed, that PSO procedure reports
much more accurate estimation of the system parameters for
the wind turbine generator system model.
3. Best value for the crossover parameter in GA is 0,6. For the
mutation rate this value is 0,5.
4. Best value for the correction factor in PSO is 2.0. For the
inertia this value is 0,8
5. GA reports a bit better efficiency, but the difference in test
time duration is very small. This is due to the fact that the main
time in carrying out the tests is spent for the simulation model
of the wind turbine and to evaluate the setting of the model 2,700 times for each of the cases. Procedures themselves take
up very little processor time.
REFERENCES
[1] Bedwani, W, A., Ismail, O. M. Genetic optimization of variable structure
PID Control System, In: ACS/IEEE International Conference on
Computer Systems and Applications, 2001, pp. 2730.
[2] Siegfried Heier, "Grid Integration of Wind Energy Conversion Systems,"
John Wiley&Sons Ltd, 1998, ISBN 0-471-97143-X
[3] Kargupta, H., Smith, R. E., System identification with evolving
polynomial networks, Proceeding of the 4th International Conference on
Genetic Algorithm, University of California, San Diego, USA, 1991, pp.
370-376.
[4] Kristinsson, K,, Dumont, G, System identification and control using
Genetic Algorithms, Ieee Transactions on Systems, Man and
Cybernetics, 1992 22 (5), pp, 10331046,
[5] Holland J.H., Adaptation in natural and artificial system, Ann Arbor, The
University of Michigan Press, 1975.
[6] http://iridia.ulb.ac.be/~mdorigo/ACO/ACO.html
[7] http://www.engr.iupui.edu/~shi/Coference/psopap4.html
[8] http://www.engr.iupui.edu/~eberhart/web/PSObook.html
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.52662
57
BIOGRAPHIES
ROSITSA KAZAKOVA was born in Belozem,
Bulgaria, in 1957. She received the M.S degree
(Automatic Control Systems) in University of Food
Technologies of Plovdiv in 1983. Since 1988 she is
an assistant professor in Technical University of
Sofia, branch Plovdiv.
She is research interest are in Artificial Intelligence,
Measurement and Automation.
STEFKA NEDELCHEVA is Professor, PhD Eng.

and from 2001 she is Head of Department
Electrical
Engineering,
Electronics
and
Automation in Faculty of Engineering and
Pedagogy- Sliven at Technical University of Sofia.
She becomes a Ch. Assistant Professor in the
Department of "Electrical Engineering" in 1990,
then Associate Professor in 1995 and Head of
Department "Electrical Engineering" (1996-1997).
Lectures on Electrical Networks, Electrical
Energy Production, Mathematical Methods in
Electrical Energy Production, Decentralized Energy Sources, Ecology
and Renewable Energy Sources, Power line Protection Design and Smart
grids.
Interests: Study and modeling modes of operation in electrical networks and
systems, application of mathematical methods in electricity, decentralized
electricity sources.
RUMEN POPOV was born in Plovdiv, Bulgaria,
in 1964. He received the M.S. degree (Automatic
Control Systems) in Technical University of Tula,
Russia, in 1990 and the Ph.D. in Technical
University of Ruse, Bulgaria in 2009.
From 1990 to 1991, he was a Research Engineer
with the Aviotechnics SA, Plovdiv at Radio
Electronic Systems department. Since 1991, he has
been an Assistant Professor with the Control
Systems Department, Technical University of
Sofia, branch Plovdiv. He became an associated
professor at the Technical University of Sofia, Bulgaria in 2011. At present hi
is an associated professor at Department of Electronics, Communications and
Information Technologies in Plovdiv University Paisii Hilendarski He is
the author of more than 62 articles. His research interests include
Measurement and Automation, Renewable Energy Systems, Sun - Tracking
systems. His main courses are in Measurement Systems, SCADA Systems,
PLC and Signals and Systems. Assoc. Prof. Popov is a member of the
Bulgarian Geothermal Association since 2011.
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING,
Copyright BAJECE
ISSN: 2147-284X
September 2016
DOI:10.17694/bajece.73922
Vol.4 No.2
58
Copyright BAJECE
ISSN: 2147-284X
September 2016
DOI:10.17694/bajece.73922
Vol.4 No.2
59
Copyright BAJECE
ISSN: 2147-284X
September 2016
DOI:10.17694/bajece.73922
Vol.4 No.2
60
Copyright BAJECE
ISSN: 2147-284X
September 2016
DOI:10.17694/bajece.73922
Vol.4 No.2
61
62
DOI: 10.17694/bajece.73922
Hybrid wind-solar reliable solution for Turkey

to meet electric demand
K. Dawood
Abstract Energy is the most important part of life, it is

essential for social and economic development. Turkey is importing
more than half of the energy from other countries to produce
electricity by thermal plants. Air pollution is also becoming quite a
big problem for Turkey due to the thermal production of the energy.
One of the most effective solution for air pollution is renewable
energy because nowadays renewable energy is environment
friendly. Turkey has many geographical location advantages one of
them is renewable energy resources. Turkey has limited fossil fuel
resources and due to this reason Turkey must shift more electricity
production to the renewable energy resources. Renewable energy
has less environment impacts as compare to the fossil fuels but still
renewable energy sources are not reliable and competitive as
compare to the fossil fuels. The present study deals with the
advantages of Hybrid renewable energy systems (Solar and Wind
energy) in Turkey.
Map of Turkey with high resources of solar-wind hybrid energy
is also presented in this paper. Integration of Wind and solar
energy into the Turkish energy will be one of the main tools in
responding to the growing electricity demand because it will
decrease the dependence of energy on single method. In addition,
non-renewable energy is also compared with the renewable energy
sources on the economical basis.
Index Terms Renewable energy , Hybrid energy in Turkey,
Solar energy, Wind Energy , Electricity generation in Turkey.
I. INTRODUCTION
NERGY is one of the crucial inputs for socio-economic

development. The rate at which energy is being consumed
by a nation often reflects the level of prosperity that it could
achieve [1]. Domestic oil and natural gas are not in sufficient
quantity in the world to support the increasing demand of
electricity. Air pollution is also becoming quite a big problem
for the world. Thermal power production of the energy is also
main reason of air pollution. One of the most effective
solution for air pollution is renewable energy because
nowadays renewable energy production is environment
friendly. Renewable energy technologies can help countries to
meet their policy goals for secure, reliable and affordable
energy.
Turkey is importing more than half of the energy from other
countries to produce electricity. Turkey has many
geographical locational advantages, one of them is renewable
energy resources. Turkey has limited fossil fuel resources and
due to this reason Turkey must have to shift more electricity
production to the renewable energy resources.
K. DAWOOD, is with Department of Electrical Engineering University of

Kocaeli, Kocaeli, Turkey, (e-mail: kamransdaud@yahoo.com).
Copyright BAJECE
ISSN: 2147-284X
Fig.1. Generation of electric power since 2011 [2]
Figure 1 shows the generation of electricity in Turkey from

2011 to 2015. Last year, more than 68 percent of total
electricity was generated by thermal power plants.
Approximately 5 percent of the electricity was generated from
renewable assets of Turkey, 26 percent of electricity generated
through hydropower plants.
Turkey is one of the richest country in the world in terms of
renewable resources. For geothermal energy Turkey is ranked
first in Europe and seventh in the world. Installed capacity of
wind energy in Turkey is increasing rapidly from 20MW in
2002 to 4503MW in 2015.
TABLE.1
TURKEY RENEWABLE ENERGY PLAN FOR 2023 [3]
Installed Power Capacity
(MW)
Electricity Generation (GWH)
2013
2023
Increment
2013
2023
Increment
Hydro
22,289
34,000
53%
59,420
91,800
54%
Wind
2,759
20,000
625%
7,558
50,000
562%
Geothermal
310
1,000
223%
1,364
5,100
274%
Solar
5,000
8,000
Biomass
224
1,000
346%
1,171
4,533
287%
Renewable
energy
Technology
II. IMPORTANCE OF RENEWABLE ENERGY IN TURKEY

In Turkey, electricity demand is increasing day by day and
Turkey cannot rely on fossil fuel because Turkey is importing
almost all of the fossil fuel from other countries. To fulfil the
demand of energy Renewable energy is the best solution due
to the insufficient resources of the non-renewable energy.
Another big problem with the thermal projects is air pollution.
Turkey is also concerned about the greenhouse gas (GHG)
emissions which are rising rapidly: in 2010, they reached
403.5 Mt CO2-eq., a two-fold increase on 1990 levels [4]. Of
this, 71% was emitted by the energy sector [5] to which
electricity generation contributed 25% or 99 Mt CO2-eq [6].
As shown in Figure-2 fossil fuels like natural gas, coal and
LNG are more than 50% of Turkeys installed capacity in
2015. Hydroelectric installed capacity in 2015 was 35.4%.
Installed capacity of wind power plant was 4503.2MW. Solar
energy share is less as compare to others but still 248.8MW
September 2016
Vol.4 No.2
capacity was installed in 2015 which is 0.3% of total installed

capacity.
63
DOI: 10.17694/bajece.73922
TABLE-3.[8]
PRODUCTION COST OF 1 MWH FROM DIFFERENT POWER SOURCES
Production cost ($/MWh)
Power source
Fig.2. Installed Capacity of 2015 in Turkey [2]
Renewable energy can be divided into two main types. First

where sun is the source; biomass, wind energy, hydro energy
and solar energy is part of that energy. Tidal energy and
geothermal energy is part of second type where sun is not
source of the energy.
The renewable energy potential of Turkey is shown in Table
2. According to this table, Solar energy potential in Turkey is
977,000 (TWh/year) from which 6105 (TWh/year) is
technically possible. However due to the high price of the
solar energy only 305 (TWh/year) is economic potential but
since few years price of the solar panels are decreasing rapidly
and it will surely increase the economic potential of solar
energy in Turkey. Wind energy technical potential is 290
TWh/year. Technical potential of hydraulic energy is 216
TWh/year and economical potential is 127.4 TWh/year.
Minimum
Average
Maximum
Conventional Coal
87.1
95.1
119
Advanced Coal
Natural Gas Conventional
Combustion Turbine
Natural Gas Advanced
Combustion Turbine
106.1
115.7
136.1
107.3
141.5
156.4
94.6
113.5
126.8
Advanced Nuclear
91.8
95.2
101
Geothermal
43.8
47.8
52.1
Biomass
90
100.5
117.4
Wind
65.6
73.6
81.6
Wind Offshore
169.5
196.9
269.8
Solar PV
97.8
125.3
193.3
Solar Thermal
174.4
239.7
382.5
Hydroelectric
69.3
83.5
107.2
Wind energy is a clean and environment friendly energy. Price

of wave and solar PV is relatively high when it is compared
with other methods.
TABLE 2.
RENEWABLE ENERGY SOURCES POTENTIAL IN TURKEY[7]
Renewable energy Source
Usage kind of
energy
Potential
Solar energy
kWh/m
1,520
Hydraulic energy
TWh/year
160
Wind energy
TWh/year
48
Geotermal energy
MWt
31,500
Biomass energy
MTEP
8.6
Biogas
MTEP
1.5-2
Fig.3. Price and Installation of solar panels [9]
Figure 3 shows the price and global installations of solar panel

from 1975 to 2015. The falling cost of solar panel per watt
caused the increment in the global installations of solar panels.
The solar energy in the world is increasing rapidly due to the
decreasing price of the solar panel. In 1975 price of a solar
panel per watt was 101$ which is decreased to the 0.61$ in
2015.
IV. RENEWABLE ENERGY IN TURKEY
III. ECONOMIC ANALYSIS OF POWER GENERATION FROM

DIFFERENT SOURCES :
Table-3 shows a 1MWh electricity levelized cost comparison
between different types of energy sources. Renewable energy
like hydroelectric, wind, geothermal and biomass energy are
not too much expensive. The 1MWh production cost of wind
energy is almost equal to the cost of 1MWh electricity
produced from coal. However, there are many advantages of
wind energy when it is compared with the conventional energy
sources.
Copyright BAJECE
ISSN: 2147-284X
A. Hydroelectric Energy
Hydroelectricity provide more than 16% of the worlds
electricity. Fossil fuels and natural gas potential is very low in
Turkey, but potential of hydropower is very high in Turkey.
Hydropower source is the second largest source to generate
electricity in Turkey, coal is the first one. In developing
country Hydroelectricity provide around 40% of the electric
energy. Hydroelectric plants are more reliable and highly
efficient. However, the dams and their large lake forms also
have major environmental and social impacts [10,11].
Turkey has significant potential of renewable energy despite
of this, the countrys energy still depends on the fossil fuels,
September 2016
Vol.4 No.2
almost 65% of the total energy produced by fossil fuels in

2015. Hydropower is one of the most common renewable
energy in the world and it also play vital role in the Turkeys
energy production. The theoretical viable hydroelectric
potential of the country has been estimated at 433 TWh/year,
nearly 1% of the total hydropower potential of the world
[12,13]. But due to the technological limitations the
hydroelectric potential almost decreases to the 50%.
In 2015, the total installed capacity of hydropower in Turkey
was 25868 MW, and total generated energy in 2015 from
hydropower was 66903.2 GWh.
B. Wind Energy
Wind energy is one of the most environment friendly and costeffective solution for the generation of electricity. Wind power
generation is increasing rapidly in the world. About 95% of
wind capacity is now located in Asia (China and India),
Europe and North America; and 90% of new capacity was
installed in those regions in 2015 (57.1 GW) [14].
Fig.4. Wind Installed Capacity in Turkey [2]
Figure 4 shows the installed capacity of wind power plants

since 2006. The installed capacity was 51 MW in 2006 and it
is increased to the 4498MW in 2015. Wind power in Turkey is
increasing tremendously but still the desired level of the
installed capacity is not reached because the wind potential is
very high in Turkey. Despite this tremendous increase, the
desired level of installed capacity is still not reached and a big
share of the economic potential has still not been used.
The technical wind energy potential is estimated to be about
114 GW of capacity in regions where the wind speed is higher
than 7.0 meters above the ground at 50 meters height [15].
C. Solar Energy
The major solar development in 2015 was the 48 % increase in
solar capacity in Asia, with 15 GW of new capacity installed
in China and another 10 GW in Japan. An increase of 8 GW in
North America also surpassed the level of new installations in
Europe (7.6 GW) for the first time [14].
DOI: 10.17694/bajece.73922
64
The installed capacity of solar energy in Turkey was just 40 in

2014 and it is increased to the 248.8MW in 2015 and total
generation of solar energy in 2015 was 307MW in Turkey.
Solar power in Turkey is increasing massively. Although solar
panel prices are decreasing but still the generation cost of solar
power is high as compare to the other power generation
methods. Solar potential is very high in Turkey, especially in
South Eastern Anatolia and Mediterranean regions solar
energy potential is very high.
The government is also encouraging expansion and the
utilization of solar energy for electricity generation. To
stimulate investment in renewables, various incentive schemes
have been introduced [16,17]. For example, renewable energy
plants with an installed capacity of 500 kW or less are exempt
from licensing obligations [18].
D. Geothermal Energy
Geothermal energy is thermal energy which is derived from
the earth's internal heat. Use of the geothermal energy is not
new and some countries have used geothermal energy for
cooking and heating for thousands of years. In 1904 first time
geothermal energy was used for electricity.
Geothermal power capacity is increased by about 600 MW
in 2015. Four countries (Italy, Mexico, Turkey and the United
States of America) accounted for almost all of this increase.
[14] Turkey is ranked seventh richest country in the world in
geothermal potential. Turkey has significant potential of
geothermal energy. In 2015, the total installed capacity of
geothermal energy was 624MW. Turkey is one of the richest
countries in the world in terms of geothermal energy
resources, with the overall potential of 31.5 GW [19].
E. Biomass and Biofuel Energy
Agricultural, forest, animal and other wastes are source of
biomass and these can be used in the production of the energy.
Biomass can be converted into biofuel as well. Biomass is
generally used as a raw product in power turbines to generate
steam. Biomass is one of the major resource from the
renewable energy, as shown in Table-2 Turkey yearly biomass
economical potential is 8.6 mtoe/year.
Turkeys domestic energy consumption accounts for about
37% of total energy consumption. Of this, about 52% is from
biomass-based fuels [20]. The potential of important biomass
energy sources and animal solid wastes of Turkey has been
determined and the potential of these as a source of fuel in
thermal generating systems to produce electricity and thermal
energy has been studied [21,22]. The available agricultural
and animal residues in Turkey have been estimated to be
roughly equal to 22 - 27% of energy consumption in Turkey
[22]. The electrical production from usable biomass (about 17
Mtoe/year) has a net impact of $4.4 billion in personal and
corporate income and represents more than 160,000 jobs [23].
Generated electricity from biomass was 1171MW in 2013 and
4533MW is the target of Turkish government till 2023.
V. HYBRID ENERGY
Fig.5. Solar energy in Turkey [2]
Copyright BAJECE
ISSN: 2147-284X
Wind power generation is not increasing rapidly in Turkey as

in the other parts of the world and main reason is variations in
wind speed. Southern and eastern areas of the country have
less wind potential as compare to the western areas of the
country. Marmara and East Mediterranean area are one from
the best wind potential in the Turkey.
September 2016
Vol.4 No.2
Integration of a large amount of renewable energy to the grid

is still one of the biggest problem for renewable energy.
Especially when the source of renewable energy is single. If
integration is not well planned, it will lead to the large scale
blackouts. Another efficient solution for that problem is hybrid
renewable energy solution. Turkey has several geographical
advantages; solar and wind energy is one from these
DOI: 10.17694/bajece.73922
65
advantages. Many areas of turkey have high wind potential

and many areas have high solar potential. Some of the areas
have high wind energy potential with high potential of solar
energy as well. Fig. 6 highlighted areas with high potential of
both solar and wind energy. Color representation is given in
table 4. anakalle, Antakya, Mardin are the regions where
both wind and solar energy resources are very high.
Fig. 6. Hybrid Energy in Turkey.
TABLE.4
Color representation of Figure 6

Wind [m/sec]
wind>3
wind>3
2.4< wind<3
2.4< wind<3
Global Horizontal Irradiation

]
irradiance>1700
Color
Red
1550<irradiance>1700
irradiance>1700
1550<irradiance>1700
Orange
Green
Yellow
Global Horizontal Irradiation is taken from [24] and it is

average annual sum of 6 years from 2004 to 2010. Wind speed
is taken from [25]. Hybrid renewable energy also has greater
economic and environment returns when it is compared with
stand-alone systems like geothermal, wind or solar systems.
There are many advantages of wind- solar hybrid system. The
system will be more reliable because there are two different
power production systems are used. Dependence on single
power system will be also minimize by using wind- solar
hybrid system which will decrease the number of storage
batteries. Another main advantage is that if one system is
down for maintenance or repairing, system can rely on the
other one, similar for the windy and sunny conditions.
VI. CONCLUSION
Wind energy, solar energy and other types of renewable
energy is important kind of sustainable energy. Demand of the
electricity is increasing day by day and Turkey cannot rely on
the fossil fuels because in Turkey these resources are limited.
However, resources of renewable energy like hydroelectric
energy, wind, solar, geothermal and biomass energy is much
higher. Renewable energy like wind and solar energy have
many advantages over other energy types but main
disadvantage of wind and solar energy is reliability. This issue
Copyright BAJECE
ISSN: 2147-284X
can be resolved by using hybrid energy. This paper also

presents the map of Turkey where we can use Wind-solar
hybrid plants. Hybrid energy make system more reliable, even
with the no wind on hybrid system we can generate energy
from solar energy.
REFERENCES
[1]
Mathew S., Wind Energy Fundamentals, Resource Analysis and

Economics, Springer, 2006.
[2] Trkiye Elektrik letim A.., www.teias.gov.tr/
[3] National Renewable Energy Action Plan for Turkey, December 2014.
http://www.eie.gov.tr/duyurular_haberler/document/National_Renewab
le_Energy_Action_For_Turkey.PDF
[4] EEA, European Environment Agency, Greenhouse Gas Data Viewer:
European Environment Agency, 2012 [Online]. Available from:
http://www.eea.europa.eu/data-and-maps/data/dataviewers/greenhouse-gases-viewer.
[5] TUIK, National Greenhouse Gas Inventory Report, 1990-2010, Turkish
Statistical Institute, Ankara, 2011.K. Elissa, Title of paper if known,
unpublished.
[6] Future Camp, Baseline Emission Calculations. Verified Carbon
Standard (VCS), Version 3, Turkey, Ankara, 2011.
[7] Usta R., Tkiye'nin yenilenebilir Enerji Stratejisi ve Politikalari, March
2015. http://www.tepav.org.tr/
[8] Levelized Cost and Levelized Avoided Cost of New Generation
Resources in the Annual Energy Outlook 2015 US DOE Annual
Energy Outlook 2015.
[9] Bloomberg
http://www.treehugger.com/renewable-energy/strikingchart-showing-solar-power-will-take-over-world.html
[10] Kaygusuz K., Hydropower potential in Turkey. Energy Sources
1999;21:5818.
[11] Kaygusuz, K. and Kaygusuz, A., Geothermal Energy in Turkey: the
Sustainable Future. Renewable and Sustainable Energy Reviews,
August
2004,
http://www.inference.phy.cam.ac.uk/sustainable/refs/geothermal/sdartic
le.pdf
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.73922
BIOGRAPHY
[12] Kilic F.C., Recent renewable energy developments, studies,
incentives in Turkey, Energy Educ. Sci. Technol. Part A 28 (1)
(2011) 37e54.
[13] Simsek H.A., Simsek N., Recent incentives for renewable energy in
Turkey, Energy Policy 63 (0) (2013) 521e530.
[14] Baris K., Kucukali S., Availibility of renewable energy sources in
Turkey: Current situation, potential, government policies and the EU
perspective, Energy Policy 42 (0) (2012) 377e391.
[15] EMRA, RE: Data on Energy Potential of Turkey, Republic of
Turkey Energy Market Regulatory Authority, Ankara, 2014
[Personel communication,15.06.2014].
[16] Demirbas A. Energy balance, energy sources, energy policy, future
developments and energy investments in Turkey. Energy Convers
Management, 2001, Vol.42, No.10, pp.1239-1258.
[17] Demirbas A. Biomass co-firing for coal-fired boilers. Energy
Exploration & Exploitation, 2003, Vol.21, No.3, pp.269-278.
[18] Balat M., The use of renewable energy sources for energy in Turkey
and potential trends. Energy Exploration & Exploitation, 2004,
Vol.22, No.4, pp.241-257.
[19] Bascetincelik A, Karaca C, Ozturk HH, Kacra M, Kaya D, Ekinci
K, et al. First progress report of exploitation of agricultural residues
in Turkey. European Commission, LIFE third countries projects,
2004e2005. LIFE 03 TCY/TR/ 000061; 2005.
[20] GHI Solar Map 2016 Solargis http://solargis.com
[21] Meteorological
Data
Processing
Department
http://www.mgm.gov.tr/
Copyright BAJECE
ISSN: 2147-284X
KAMRAN DAWOOD was born in Karachi, Pakistan

in 1989. He received his B.E. Electrical engineering
degree from Hamdard University, Karachi, Pakistan,
in 2013. Hamdard University also awarded him gold
medal for securing first position in Electrical
Engineering. Currently he is doing M.Sc. degree in
Electrical engineering from Kocaeli University, Izmit,
Turkey. His current research interests involve
renewable energy, power quality, smart grids and
electric vehicles.
September 2016
Vol.4 No.2
66
DOI: 10.17694/bajece.41977
Hardware implementation of auto switching

and light intensity control of LED lamps
S. Singh and S. Bhullar
Abstract This paper presents energy efficient lighting

control architecture for street lights. The system utilizes auto
intensity control technology to be implemented on street lights
to make it a smart street lighting system. The proposed system
consists of LED Lamps, infrared transmitter and receiver, NE555 timing IC, SPDT relays, PV panel, etc. Intensity of street
lamps is controlled depending upon the movement of vehicles
on the road. When any vehicle passes on the road, increases the
light intensity of lamps for a few seconds and automatically set
in previous position when a vehicle passes out. Also manpower
is totally eliminated by the use of auto switching of street light.
This is the best and cheapest method for saving energy and also
for possible reducing lighting pollution. The PV system is used
for charging the batteries, the power of which is used during
the night and grid source is also used as a backup of our system
when sometimes PV Off-Grid goes faulty.
Index Terms Energy efficient system, Smart Street lighting,

LED, HPS, CRI, NO, NC, auto light intensity control, PV offGrid, Light pollution.
I. INTRODUCTION
HE main scope of outdoor lighting is the extension of

an acceptable human life quality consists of i.e.
accessibility of social activities, traffic safety during night,
crime prevention and human behavior during dark period of
the day. The main purpose of road lighting is to make
people, vehicles and objects on the road visible.
In these days, HPS Lamp is widely used for
street lighting system which produces yellow light with low
CRI, short life up to 5000 hrs, contains a lead element. The
main limitation is that it takes some time for initial starting.
It consumes more power as compared to other lighting
lamps. [1]
Todays latest technology LED Lamps which have high
CRI nearly greater than 80, produce white cool light, long
life up to 50,000 hrs, contains no lead element, less ignition
time, less maintenance and no effect of switching on its life.
It consume less power as compared to HPS, but its initial
cost is high which is one of the limitation of LED [2]. We
can save a lot of energy and reduce tons of carbon dioxide
emissions per year. Conventional lighting systems are
working on bi-level approach either ON or OFF. When ON,
it operates on its rated power and there is no control to
adjust its light intensity according to the availability of
traffic on roads. The light intensity control system is
introduced to reduce energy consumption inherent to
S. SINGH, is with Thapar University, Department of Electrical and
Instrumentation,
Patiala,
India
(e-mail:
satwindersinghsekhon@hotmail.com).
S. BHULLAR, is with Thapar University, Department of Electrical and
Instrumentation, Patiala, India (e-mail: suman.bhullar@gmail.com).
conventional lighting system. The PV panel is used to

charge the battery whose power is used during the night for
street lights[3]. In this work system power is used from both
grid source as well as from a PV system. Analysis between
HPS and LED lamp.
Objective: The objective of the work is to control energy
consumption, reduction of Co2 emissions and to obtained
sustainable system.
II. SYSTEM ARCHITECTURE
The system combines advanced and traditional controlling
features. The proposed system minimizes the energy
consumption for the benefit of users and environment
friendly.
The system comprises of Infrared transmitter, infrared
receiver, PV Panel, transformer, NE-555 timing IC, SPDT
relays, LED Lamps, rectifier and filter unit, LDR, 7805
voltage regulator, BC 547 transistor.
A. Infrared transmitter
Infrared transmitter used white infrared LED, timing IC in a
stable mode where it can operate as an oscillator and
produce 38 KHz signal [4] which continually fall on the
receiver.
B. Infrared receiver
TSOP 382 sensor is used as a receiver in the hardware
model project, which is placed on opposite side of
transmitter near the street light tower. Output of TSOP382
will be LOW when no signals falls on it and the output will
be HIGH when 38 KHz infrared rays falls on it [5].
C. PV Panel
12V, 12W PV panel is used for charging the 12V, 2.5 Ah
batteries that power are used at night for street lights.
D. Transformer
It is used stepping down the 220V into 12V to operate the
entire circuit.
E. NE 555 Timing IC
This IC is used in two modes in this circuit, one is a stable
and another is mono-stable mode. A stable mode is used in
transmitter circuit and Mono-stable mode is used for timing
delay that depends upon the sensitivity of traffic by a sensor
on the road. [6]
F. SPDT Relay
It is an electromagnetic device which is used to isolate two
circuits electrically and connect them magnetically. They
----------------------------------------------------------------------------------------------------------------------------------------------------------------------Copyright BAJECE
ISSN: 2147-284X
September 2016 Vol.4 No.2
67
are very useful devices and allow one circuit to switch

another one while they are completely separate. They are
often used to interface an electronic circuit to an electrical
circuit which works at very high voltage. When the relay is
energized they are working in NC mode and during deenergized in NO mode and COM pin is always connected
with a 5V supply [7].
DOI: 10.17694/bajece.41977
III. CIRCUIT DIAGRAM
G. LED Lamps
They are used for lighting during the night in street lights. In
this model 8 LEDs are used which are connected in parallel
across supply and each having a rating of 1W.
H. Rectifier and filter unit
Full wave rectifier is used to convert the ac output of the
transformer into DC. The output of rectifier contains ripples
which can be further removed by capacitor as a filter unit
and hence pure DC is obtained. [8]
. Light Dependent Resistor
It is a device which offers resistance in response to the
ambient light. Its resistance decreases when light falls on it
and vice versa. In the absence of light, LDR exhibits a
resistance of the order of mega-ohms, which decrease too
few hundred ohms in the presence of light [9]. It can acts as
a sensor, since a varying voltage drop can be obtained in
accordance with the varying light.
Fig. 1. Auto light intensity control of LED Lamp
A. FLOW CHART
J. 7805 voltage regulator IC

It is used in various electronic circuits where we need 5V
supply. An IC is designed as fixed voltage regulator and
with adequate heat sinking; it can deliver output current in
excess of 1A. [10]
K. BC 547 transistor
The transistor is 3 pin devices which are used as a switch in
hardware model.
L. Battery and grid source interface through relay
based circuit
We use the power form both grid sources as well as from a
battery. In cloudy days or during winter when sun light is
not available for many days or due to any other reason
batteries are not charged properly, then we can use grid
source.
An automatic interface circuit between grid source and
battery is designed. 12V relay is used which is excited
through the secondary of transformer but that is ac which is
convert into dc by the use of diode. Capacitor is used to
remove the ripples [11]. When the grid is supplying power
then the relay is excited and cut off the supply from battery
side. When power from grid side is switched off, then relay
de-energized and the battery is used as a source for street
lighting.
Fig. 2: Auto light intensity control of LED Lamp
The above flow chart explains the operation of controlling

of intensity of LED Lamps. The transmitter generates 38
KHz signal which is falling on the receiver [12]. When there
is no traffic on the road, then transmitter signal fall on the
receiver continuously and the output of the receiver will be
high then relay 1 is energized and it starts operating in NC
mode. During this period, the power to LED Lamps by
adding resistance in series is supplied through second relay
from COM pin which is connected with 5V and LED glows
in dim mode.
ISSN: 2147-284X
68
When any vehicles pass through the road, then the receiver
detects it and its output goes low for that instant only relay 1
gets de-energized and starts operating in NO mode.
When relay 1 starts
operating in NO mode, and then timing IC is also triggered.
It starts working as a short pulse generator [13]. With the
output of timing IC, relay 2 gets energized. The input supply
to the infrared receiver is fed from NO mode of the relay.
So, when relay 2 is energized and starts working in NC
mode, at that time, this supply also cuts off and current in
parallel again fed to LED lamps which increase its intensity
and starts its operation in full mode. The time interval of full
mode depends upon NE 555 timing IC. After a few seconds,
the LED Lamps again starting glow in dim mode when a
vehicle passes out. In this way, the intensity of LED lamps
is controlled which can save energy and reduces light
pollution.
IV.
DOI: 10.17694/bajece.41977
output goes HIGH and the relay gets energized. It starts

operating in NC mode [16]. Hence, the power supply from
the COM point of relay gets connected with street light. As
a result it will switch ON during the night. Next day when
the sun rises, then light will turn OFF automatically.
A. FLOW CHART
AUTO SWITCHING ON STREET LIGHT
Auto switching of the street light is used to eliminate the

manpower. By using the auto switching system of Smart
Street lighting energy consumption can be reduced because
manually operates lighting system are not switched ON
earlier before sunset and also not switched OFF properly
even the sunlight comes[14]. By implementation of auto
switching system, maintenance and energy cost can be
reduced and community satisfaction will be higher.
Fig. 4: Auto switching on the street light
V. WAVEFORMS OF CIRCUIT COMPONENTS
Fig. 3: Auto switching of Smart Street light
Fig. 5: Output waveform of NE555 in transmitter circuit
During the day: The sun rays continue falling on LDR and
its resistance decreases, which results in an increase of the
voltage at pin 2 of the IC 555. IC 555 has got comparator
inbuilt, which compares between the input voltage from
pin2 and 1/3rd of the power supply voltage [15]. The input
rises above 1/3rd, output is set LOW and the relay gets deenergized. During the de - energized mode, the relay will
operate in NO mode; hence there is no supply in NC mode.
So, the street light will not turn ON at day time.
During the night: At night time, when no light falls on
LDR i.e. its resistance increases, which results in a decrease
of the voltage at pin 2 of the IC 555, then the timing IC
Fig. 6: Infrared LED waveform in dim mode
ISSN: 2147-284X
69
DOI: 10.17694/bajece.41977
View of implemented system model in different modes
Fig. 7: Infrared LED waveform in Full mode
Fig. 11: Normal view
Fig. 8: Transformer output waveform
Fig. 12: LED Lamp glow in Dim Mode
Fig. 9: Rectifier output waveform
Fig. 13: LED Lamp glow in Full mode
Fig. 10: 7805 transistor output waveform
VI. SYSTEM IMPLEMENTATION

Auto switching and light intensity control of LED Lamps
aim to modernize infrastructure of traditional wired system,
currently exercised for outdoor lighting systems. This
method has low cost, less maintenance and can be
implemented easily. [17]
Fig. 14. Complete Hardware Setup
ISSN: 2147-284X
70
VII. POWER CONSUMPTION BY LED LAMPS

Load current consumed by LED Lamps during Dim mode =
0.12A.
Load current consumed by LED Lamps during Full mode =
0.26A.
Battery voltage = 12V.
Power = VI.
So, Power consumed during Dim mode = 120.12A =
1.44W.
Power consumed during Full mode = 120.26A = 3.12W.
So, from the above calculations it shows that we can save
energy more than 50% and reduce Co 2 emissions, if we use
this model.
VIII. CONCLUSIONS
This paper is aimed at designing and implementation of an
automatic street lighting system where in the street lights
that are not required through the night can be dimmed.
Additionally, the ambiance of light is checked and lights are
turned ON when it is dark and turned OFF during the day.
The proposed system is designed to eliminate the manpower
and reduce energy consumption of outdoor premises and
demonstrated by developing a prototype to control street
lights. The system reduces the energy consumption for the
benefit of user and environment friendly [18].
IX. ACKNOWLEDGEMENT
The author would like to acknowledge Faculty of Electrical
and Instrumentation Engineering, Thapar University, Patiala
in supporting this work.
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
REFERENCES
Sanjana, Solar Powered traffic sensitive automated LED street
lighting system, M.E Thesis, Dec 2012.
W.L. Chee, Modular Intelligent Control System, B.E Thesis, May
2011.
M. Ali., M. Orabi, E. Abdelkarim, J.A.A. Qahouq, A.E. Aroudi,
"Design and development of energy-free solar street LED light
system," IEEE PES Conference on Innovative Smart Grid
Technologies - Middle East, pp. 1- 7, 17-20 Dec. 2011.
A. Sanjana, Solar Powered traffic sensitive automated LED street
lighting system, M.E Thesis, Dec 2012.
M. Castro, A.J. Jara, A.F.G. Skarmeta, "Smart Lighting Solutions for
Smart Cities," 27th International Conference on Advanced
Information Networking and Applications Workshops, pp.13741379, 25-28 March 2013.
G.M. Ereu, J.O. Mantilla, "A Methodology to Determine Electrical
Energy Consumption in Street Lighting Systems," IEEE/PES on
Transmission & Distribution Conference and Exposition: Latin
America, pp. 1-5, 15-18 Aug. 2006.
H. Rohaida, O. Zulkifli, Automatic Street Lighting System for
Energy Efficiency based on Low Cost Microcontroller, IJSSST,
Vol.13, No.1, 2012.
J.D.Lee, K.Y. Nam, S.H. Jeong, S.B. Choi, H.S. Ryoo, D.K. Kim,
"Development of Zigbee based Street Light Control System," IEEE
PES on Power Systems Conference and Exposition, pp. 2236- 2240,
Oct. 29-Nov. 1 2006.
L. Lian, L. Li "Wireless dimming system for LED Street lamp based
on Zigbee and GPRS," 3rd International Conference on System
Science, Engineering Design and Manufacturing Informatization,
vol. no. 2, pp. 100-102, 20-21 Oct. 2012.
D.V. Pushpa Latha, Simulation of PLC based Smart Street Lighting
Control using LDR, International Journal of Latest Trends in
Engineering and Technology, Vol. 2 Issue, July 2013.
M. Reinhard, Reiner Andreas, An energy efficient pedestrian aware
Smart Street Lighting system, International Journal of Pervasive
Computing and Communications, Vol.7, No.2, pp.147-161, 2011.
R.P. Pantoni, D. Brandao, "A geocast routing algorithm intended for
street lighting system based on wireless sensor networks," 9th
IEEE/IAS International Conference on Industry Applications, pp. 16, 8-10 Nov. 2010.
M. Popa and A. Marcu, A Solution for Street Lighting in Smart
Cities, Carpathian Journal of Electronics and Computer
Engineering, Vol. 5, pp. 91-96, 2012.
DOI: 10.17694/bajece.41977
[14] R. Priyasree, R. Kauser, Automatic Street Light Intensity Control

and Road Safety Module Using Embedded System, International
Conference on Computing and Control Engineering, 12-13 April,
2012.
[15] A.A. Siddiqui, A.W. Ahmad, H.K. Yang, C. Lee, "ZigBee based
energy efficient outdoor lighting control system," 14th International
Conference on Advanced Communication Technology, pp.916-919,
19-22 Feb. 2012.
[16] M. Ali, M. Orabi, M.E. Ahmed, A. El-Aroudi, A single stage SEPIC
PFC converter for LED street lighting applications, Power and
Energy (PECon), 2010 IEEE International Conference on, Nov. 29
2010-Dec. 1 2010, pp.501-506.
[17] Yoshiura, N.; Fujii, Y.; Ohta, N., "Smart street light system looking
like usual streetlights based on sensor networks," Communications
and Information Technologies, 2013 13th International Symposium
on, pp. 633,637, 4-6 Sept. 2013.
[18] N. Zotos, C. Stergiopoulos, K. Anastasopoulos, G. Bogdos, E. Pallis,
C. Skianis, "Case study of a dimmable outdoor lighting system with
intelligent management and remote control," Telecommunications
and Multimedia (TEMU), 2012 International Conference on, pp.
43,48, July 30 2012-Aug. 1 2012.
BIOGRAPHIES
SATWINDER SINGH was born in 1988.
He received his Diploma in Electronics and
Telecommunications Engg. From PSBTE &
IT Chandigarh in 2008 and Bachelor Degree
in Electrical and Electronics Engg. From
Punjab Technical University, Jalandhar in
2012. Recently, he is completed Masters
degree in Power Systems from Electrical and
Instrumentation Department at Thapar
University, Patiala and Researched on the
topic of the Smart Street Lighting System.
SUMAN BHULLAR was born in 1976. She

received her Bachelor degree from Punjab
University, Chandigarh and Masters Degree in
Electrical Engineering from Punjab Technical
University, Jalandhar in 1997 and 2001
respectively. Presently, she is an Assistant
Professor and Pursuing Ph.D in Electrical and
Instrumentation
Department,
Thapar
University, Patiala.
ISSN: 2147-284X
71
DOI: 10.17694/bajece.61693
72
Aeronautical telecommunication simulator

D.K. Nedev, and K.D. Nedev
AbstractAeronautical telecommunication simulator is a

product specially made for education and training of standard
aviation phraseology and flight instruments. It consists of two
separate devices: a pilots radio and navigation panel based on
Pilatus PC-9M and an ATCs radio panel. They communicate with
each other via FM frequency radio signals. The hardware part of
the system is based on Arduino Mega 2560 and Arduino Uno
microcontrollers.
Index Terms-aeronautical, telecommunication, phraseology,
simulator, Arduino
I.
Each box consists of two frequencies: active and standby,

which can be switched by a push button located next to the
box. This device can transmit radio signals on two different
frequencies if 1-2 mode is turned on. In the same time, it
can receive three (two VHF and one UHF) signals. It is also
equipped with seven rotary encoders with push buttons for
adjusting volume of the incoming signals of each box. With
the push button we can choose the active box (the box we are
transmitting from).
INTRODUCTION
EPORTS claim that air transport is the safeties transport

in the world. However, aviation accidents do happen
occasionally and harm human life. They are unpredictable,
unpreventable and often cause loss of life. A huge percent of
all aviation accidents are caused by misunderstanding in
communication between pilots and ATCs. Even the worst air
disaster of all times - Tenerife airport disaster resulted due to
miscommunication. Therefor understanding and practicing
aeronautical telecommunications is essential for the flight
safety.
The simulator is a collection of hardware and software
particularly designed for training aeronautical phraseology
between pilots and controllers. It consists of two separate
devices: a pilots radio and navigation panel based on Pilatus
PC-9M aircraft and an ATCs radio panel. The unlimited
number of panels can be added for each exercise. For
example: two pilots and three controllers can practice together
in real time. Devices communicate with each other via FM
frequency radio signals. In addition, a pilots Audio Radio
Control Display Unit (ARCDU) from PC-9M can be
connected to a flight simulator software such as Microsoft
Flight Simulator X and X-plane.
a)
II. PILOTS RADIO AND NAVIGATION PANEL

It is based on Pilatus PC-9m aircraft. It has seven separated
boxes. Two of them control VHR radio stations, the other
three control navigation stations and there is also one for UHF
communication and one transponder identification.
D. NEDEV, works in University of Plovdiv, Technological Faculty, Smolyan,

Bulgaria, (e-mail: dim_nedev@abv.bg)
K. NEDEV, works in Graf Ignatievo Air Force Base, Bulgaria (e-mail:
koko_nedev@abv.bg)
Copyright BAJECE
ISSN: 2147-284X
b)
Fig.1.ARCDU panel from PC-9M & ARCDU simulator
On the right bottom part of the panel there is a frequency

dual rotary encoder which is used for adjusting the standby
frequency of each box. Using the PGE button main two
displayed pages of the device can be switched.
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.61693
73
The radio transmitter and receiver are connected to extension

board. The wiring diagram is shown on Fig.3.
IV. SOFTWARE
Software part has been created with Arduino IDE. The code
consists of more than two hundred lines. First main function
is called drawHomeScreen (). It uses UTFT library to
visualize home page.
Fig. 2. ARCDU page 2 & ARCDU simulators display
III. HARDWARE
Fig. 4. Arduino ARCDU code
Fig. 3. Wiring diagram
Hardware part is based on Arduino Mega 2560

microcontroller, a 3.2 TFT LCD screen and 12V power
supply unit. It also incorporates with 7 encoders, two
potentiometers, one switch and sixteen push buttons. Six of
the encoders control volume level of each box, seventh is dual
and adjusts standby frequency of each box. All hardware parts
are connected to microcontrollers digital pins except for the
potentiometers which are connected to analog once. The first
twenty pins from the Arduino microcontroller are used for the
rotary encoders. The next twenty pins link microcontroller
with the TFT display. The pins from 42 till 53 are used for
controlling the push buttons. All these pins use digital serial
communication. There are also two analog potentiometers
connected to analog pins A0 and A01. The first one controls
the brightness of the TFT display. The other controls the IC.
Copyright BAJECE
ISSN: 2147-284X
Fig. 5. Arduino ARCDU main function
The code of the program includes more than twenty other

custom functions. It also has two standard functions for each
Arduino program: void setup () and void loop (). The first one
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.61693
is executed only once following program startup. It is used to

define microcontrollers pins as INPUT or OUTPUT. The
second standard function void loop works all the time when
the device is turned on. It constantly monitors the status of all
encoders, push buttons and potentiometers. The code has
modular structure and easily can be modified and updated.
ACKNOWLEDGEMENT
I would like to thank to Dr. Remigiusz Pielacha
Aeronautical Telecommunication lecturer in Polish Air Force
Academy for providing materials in the area of aviation
telecommunication.
V. ATC RADIO PANEL
REFERENCES
It has one 2x8 chars LCD display, two rotary encoders and a
switch button. It is also equipped with FM radio receiver and
transmitter. It can receive and transmit signal to only one
frequency.
74
[1] Krushkov Hr., ++ programming, Plovdiv, 2006

[2] Kenarov N., IC microcontrollers, Varna, 2006
[3] Kenarov N., Microcontrollers in exercises, Varna, 2006
[4] http://howtomechatronics.com/
[5] www.arduino.cc
[6] Jeremy Blum, Exploring Arduino, 2016
[7] Kraig Mitzner, Complete PCB Design Using OrCAD
[8] James M. Kirkpatrick, Electronic Drafting and PCB, Second edition
BIOGRAPHIES
DIMITAR K. NEDEV was born in Smolyan,
Bulgaria. He received the B.S. and M.S degrees in
Microelectronics from "Angel Kanchev" University of
Ruse, Ruse, Bulgaria. Since 2015, he is a lecturer in
Paisii Hilendarski University of Plovdiv, Bulgaria.
Her research interests are in electronics and particularly
in designing PCBs.
a)
KOSTADIN D. NEDEV was born in Smolyan,

Bulgaria. He received the B.S. degree in Aviation from
National Military University, Aviation Faculty, Dolna
Mitropolia, Bulgaria. He has passed also ATPL course
in Polish Air Force Academy, Deblin, Poland. Since
2016 he works as a pilot in Bulgarian Air Force.
b)
Fig. 6. ATCs panel
VI. CONCLUSION
According
to
our
experiments
Aeronautical
telecommunication simulator is a very useful training system
for pilots and ATCs. It can be used from beginner and
experienced pilots and controllers. The system is part of a new
aviation education system used in Bulgarian Air Force, which
is going to be announced in February 2017.
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.14351
75
A New Approach of Browns Double

Exponential Smoothing Method in Time Series
Analysis
S. Hansun
AbstractDouble Exponential Smoothing is an improvement of

Simple Exponential Smoothing, also known as Exponential
Moving Average, which does the exponential filter process twice.
Its usually been used to predict the future data in time series
analysis, where there is a trend in the data. In this paper, we aim
to introduce a new approach of Browns Double Exponential
Smoothing in time series analysis. The new approach will combine
the calculation of weighting factor in Weighted Moving Average
and implement the results with Browns Double Exponential
Smoothing method. The proposed method will be tested on
Jakarta Stock Exchange (JKSE) composite index data. The result
of the proposed method shows a promising result in this work.
Index TermsBrowns Double Exponential Smoothing, JKSE
composite index, time series analysis, Weighted Moving Average.
I. INTRODUCTION
The simplest type of moving average is Simple Moving

Average (SMA), where each point in time series data is
weighted the same, regardless of where it occurs in the
sequence. Weighted Moving Average (WMA) is an
improvement of SMA, which gives a weighting factor for each
point in time series data. Another type of moving average is
Exponential Moving Average (EMA) which actually is a
variation of WMA that used exponential number as the basis in
forming weighting factors in time series analysis.
Hansun had introduced a new approach of moving average
method which combine the WMA and EMA methods to
forecast the future data in a time series data [14]. In this
research, we will further develop the hybrid method introduced
by Hansun, by modifying and combining the WMA method
with Browns Double Exponential Smoothing method. The
proposed method will be implemented on Jakarta Stock
Exchange composite index data and be compared with other
moving average methods.
TIME SERIES is a set of regular time-ordered

observations of a quantitative characteristic of an
individual or collective phenomenon taken at successive, in
most cases equidistant, periods or points of time [1]. Mainly,
the usage of time series models can be divided into two, i.e. to
obtain an understanding of the underlying forces and structures
that produced the observed data, and to fit a model and proceed
to forecasting, monitoring, or even feedback and forward
control [2]. In order to understand the characteristics of a time
series, many researchers have adopted and developed time
series analysis methods with the final aim to find a pattern
which can be used to forecast future event or data [3, 4, 5].
One of time series method that has been used widely by people
is the moving average method. Although many other soft
computing methods have been developed recently, as we can
find in the works of Stevenson and Porter [6], Hansun [7, 8],
and Popoola [9,10], the moving average method still considered
as the best method by many people due to its easiness,
objectiveness, robustness, and usefulness [11, 12].
There are various kinds of moving average methods, but their
underlying purpose remain the same, that is to track the trend
determination of the given time series data [11, 13].
S. HANSUN, is with Department of Computer Science Universitas Multimedia

Nusantara, Tangerang, Indonesia, (e-mail: hansun@umn.ac.id).
Copyright BAJECE
ISSN: 2147-284X
II. WEIGHTED MOVING AVERAGE AND BROWNS DOUBLE

EXPONENTIAL SMOOTHING
The proposed method in this research combines the Weighted
Moving Average and Browns Double Exponential Smoothing
which is one of the well-known exponential smoothing methods
in time series analysis.
A. Weighted Moving Average (WMA)
A Weighted Moving Average (WMA) is an improvement form
of Simple Moving Average (SMA). It gives a greater weight to
more recent data than the older ones. The weighting factors are
calculated from the sum of days used in time series data, also
known as the sum of digits [12]. The formula used can be
described as below, where refers to the period or span number
of forecasting formula and refers to the value of time series
data at point [15].
=
+(1)1 ++2(+2) +(+1)

+(1)++2+1
(1)
B. Browns Double Exponential Smoothing

Browns Double Exponential Smoothing, also known as
Browns Linear Exponential Smoothing, is one type of double
exponential smoothing which uses two different smoothed
September 2016
Vol.4 No.2
series that are centered at different points in time [16]. The

formula is based on an extrapolation of a line through the two
centers and can be described as below [16]. Let denote the
single-smoothed series obtained by applying simple
exponential smoothing to series and denote the doublesmoothed series obtained by applying simple exponential
smoothing (using the same constant smoothing factor, ) to
series :
0 = 0
0 = 0
= + (1 )1
)1
(3)
(4)
where is the estimated level at time , and is the estimated

trend at time ,
= 2
=
( )
(5)
(6)
The proposed method will be called as Browns Weighted

Exponential Moving Average (B-WEMA), since its quite
similar to the Weighted Exponential Moving Average (WEMA)
method introduced by Hansun [14]. The main difference is on
the methods been used. In [14], WEMA combine the Weighted
Moving Average (WMA) and Exponential Moving Average
(EMA) methods, while in this research we will combine the
WMA with Browns Double Exponential Smoothing, which is
an improvement of EMA.
The algorithms procedures of the proposed method can be
written as follow:
1.
Calculate the base value, , using Eq. (1) for a given time
series data and periods
2.
Using the base value obtained, calculate the forecasting

value using Eq. (2) (6), whereas
1
= 1
=
3.
IV. FORECAST ERROR MEASUREMENT
A. Mean Square Error (MSE)

Mean Square Error (MSE) denotes the average of the square of
error sum between the forecasted data and the actual data. The
formula can be written as follows [14],
=1
(8)
where denotes the number of data, denotes the forecasting

error from . Here, is the actual data and is the
forecasted data.
B. Mean Absolute Percentage Error (MAPE)
Mean Absolute Percentage Error (MAPE) value gives us an
indication about how much the average of absolute error of the
forecasted data compare to the actual data, and denotes by the
formula [14],
III. BROWNS WEIGHTED EXPONENTIAL MOVING AVERAGE

(B-WEMA)
76
To compare and measure the accuracy and robustness of the

proposed method against other moving average method, we use
two common measurement tools, i.e. Mean Square Error (MSE)
and Mean Absolute Percentage Error (MAPE).
(2)
=
+ (1
Then, the forecast for + , for any > 1, is given by:
+ = + ,
DOI: 10.17694/bajece.14351
=1| |
100
(9)
where denotes the number of data, and denotes the

forecasting error from . The actual data is denoted by
, while denote the forecasted data.
V. EXPERIMENTAL RESULTS AND ANALYSIS
To test the accuracy and robustness of the proposed method, we
will try to implement it in Jakarta Stock Exchange (JKSE)
composite index data. The number of data been used were 100
JKSE data, taken monthly from April 2007 to July 2015 [17].
The length or span data (i.e. the specified number of data points
to be used taken backward from the most recent data) and the
initial data to start with can be chosen by the user. Mean Square
Error (MSE) and Mean Absolute Percentage Error (MAPE) will
be used to calculate and compare the accuracy and robustness
of the proposed method against the other two moving average
method, i.e. Weighted Moving Average (WMA) and Browns
Double Exponential Smoothing.
(7)
Return to step 1 until each data point in the time series data
given ended.
Fig.1. Interface of the system
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
Fig. 1 shows the interface of the system. User can choose any
value for the initial data to start with and span data as described
above. As shown on Fig. 1, the initial data been used in this
DOI: 10.17694/bajece.14351
77
experiment is 29 and the span data been used is 5, which means

that the calculation will be started from the 30th data period
considering the last 5 data taken successively.
Fig.2. Weighted Moving Average Forecasting Results
Fig.3. Browns Double Exponential Smoothing Forecasting Results
Fig.4. Browns Weighted Exponential Moving Average Forecasting Results
The graph of forecasted data based on Weighted Moving

Average (WMA), Browns Double Exponential Smoothing (BDES), and Browns Weighted Exponential Moving Average
(B-WEMA) are shown on Fig. 2, Fig. 3, and Fig. 4
consecutively. On the figures, the actual data is denoted by the
blue line, while the forecasted data is denoted by the red line
with a triangle mark on each forecasted point.
Copyright BAJECE
ISSN: 2147-284X
TABLE I
MSE and MAPE values for each method
MA method
MSE
MAPE
WMA
72435.785437873
5.689034692941
B-DES
36036.961439658
3.6042411277246
B-WEMA
30805.878289729
3.6001241345362
Table 1 shows us the different MSE and MAPE values for

each method. As can be seen, the MSE and MAPE values of the
September 2016
Vol.4 No.2
proposed method are the smallest among the other methods,

which means that the proposed method can be used as a better
forecasting tool in time series analysis.
VI. CONCLUSION
In this paper, a new approach of moving average method has
been introduced. The proposed method combines the basic
formula of Weighted Moving Average (WMA) to get a base
value, and then use it to get the forecasted value using Browns
Double Exponential Smoothing formula.
The experimental result on 100 Jakarta Stock Exchange
(JKSE) composite index data shows a promising result.
Compare to Weighted Moving Average and Browns Double
Exponential Smoothing, the proposed method excels in the
accuracy and robustness of forecasted data, as can be seen from
the small Mean Square Error (MSE) and Mean Absolute
Percentage Error (MAPE) values.
As the future works, we can try to take a more comprehensive
study to analyze the advantages and disadvantages of the
proposed method compare to other hybrid moving average
method, such as Weighted Exponential Moving Average
(WEMA). Another study to combine other moving average
methods, such as Holt-Winters Triple Exponential Smoothing
can also be taken.
DOI: 10.17694/bajece.14351
78
accessed on August 19th, 2015.

[14] S. Hansun, A new approach of moving average method in time series
analysis, Proc. of the 2013 IEEE International Conference on New
Media, November 2013.
[15] incrediblecharts,
Weighted
moving
average,
https://www.incrediblecharts.com/indicators/weighted_moving_average.
php, last accessed on August 19th, 2015.
[16] R. Nau, Moving average and exponential smoothing models,
http://people.duke.edu/~rnau/411avg.htm, last accessed on August 19th,
2015.
[17] Yahoo!
Finance,
http://finance.yahoo.com/q/hp?s=%5EJKSE
&a=03&b=1&c=2007&d=06&e=7&f=2015&g=m, last accessed on July
20th, 2015.
BIOGRAPHY
SENG HANSUN Tangerang, Indonesia. He received
the S.Si. degree in Mathematics from Universitas
Gadjah Mada, Yogyakarta, in 2008 and M.Cs.
degrees in Computer Science from the same
university, Universitas Gadjah Mada, Yogyakarta, in
2011.
Since 2011, he has been a Lecturer with the
Computer
Science
Department,
Universitas
Multimedia Nusantara, Tangerang. He is the author of
two books and 26 articles. His research interests
include computational science, soft computing methods, internet and mobile
technology.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
OECD:
Glossary
of
Statistical
Terms,
https://stats.oecd.org/glossary/detail.asp?ID=2708, last accessed on
August 19th, 2015.
NIST/SEMATECH
e-Handbook
of
Statistical
Methods,
http://www.itl.nist.gov/div898/handbook/pmc/section4/pmc41.htm, last
accessed on August 19th, 2015.
Subanar and Suhartono, Wavelet Neural Networks untuk Peramalan Data
Time Series Finansial, Program Penelitian Ilmu Dasar Perguruan Tinggi,
Yogyakarta: FMIPA UGM, 2009.
Boediono and W. Koster, Teori dan Aplikasi Statistika dan Probabilitas,
Bandung: PT. Remaja Rosdakarya, 2001.
B. Render, R.M. Stair Jr., and M.E. Hanna, Quantitative Analysis for
Management, 8th ed., New Jersey: Pearson Education, Inc., 2003.
M. Stevenson and J.E. Porter, Fuzzy time series forecasting using
percentage change as the universe of discourse, World Academy of
Science, Engineering and Technology, vol. 27, no. 55, pp. 154-157, 2009.
S. Hansun, Peramalan data IHSG menggunakan fuzzy time series,
Indonesian Journal of Computing and Cybernetic Systems (IJCCS), vol.
6, no. 2, pp. 79-88, July 2012.
S. Hansun, Jakarta stock exchange forecasting using backpropagation
neural networks, Proc. of the 2013 IEEE International Conference on
Electronics Technology and Industrial Development, October 2013.
A. Popoola, S. Ahmad, and K. Ahmad, A fuzzy-wavelet method for
analyzing non-stationary time series, Proc. of the 5th International
Conference on Recent Advances in Soft Computing RASC2004, United
Kingdom: Nottingham, 2004, pp. 231-236.
A.O. Popoola, Fuzzy-wavelet method for time series analysis,
dissertation, Department of Computing, School of Electronics and
Physical Sciences, Surrey: University of Surrey, 2007.
C.
Murphy,
Moving
averages,
http://www.investopedia.com/university/movingaverage/, last accessed
on August 19th, 2015.
S. Dash, A comparative study of moving averages: simple, weighted, and
exponential,
http://www.tradestation.com/education/labs/analysisconcepts/a-comparative-study-of-moving-averages, last accessed on
August 19th, 2015.
N.E. Hwa, Different uses of moving average (MA),
http://www.chartnexus.com/learning/static/pulses_apr2007.pdf,
last
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.43067
79
Automatic Speaker Gender Identification for the

German Language
C. Bakir
Abstract Authentication systems necessitate transmission,

design and classification of biometric data in a secure manner.
Moreover, in voice process of biometric can be obtained
successful results by determining gender of speaker. In this
study, the aim was to designed system taking German sound
forms and properties for automatic recognition gender of
speaker. Approximately 2658 German voice samples of words
and clauses with differing lengths have been collected from 50
males and 50 females. This voice samples includes more than one
word as a word. Features of these voice samples have been
obtained using MFCC (Mel Frequency Cepstral Coefficients).
Feature vectors of the voice samples obtained have been trained
with such methods as Hidden Markov Model, Dynamic Time
Warping and Artifical Neural Network. In the test phase, gender
of a given voice sample has been identified taking the trained
voice samples into consideration. Results and performances of
the algorithms employed in the study for classification have been
also demonstrated in a comparative manner.
Index TermsSpeaker Gender Recognition System; Hidden

Markov Model; Dynamic Time Warping; Artificial Neural
Networks.
I. INTRODUCTION
OWADAYS, the security problems have started to pop
out along with the development of the technology. The

security issue comes at the helm of these problems. Especially,
the biometric systems, such as identity authentications, form
the most significant part of the security issue. Thus, forensic
sound examinations of the audio records, which are the subject
of various crimes, are required. Some studies have been made
in order to prevent the leakage of information, which belong to
the persons, particularly in commercial transaction to other
persons. Hand Writing Recognition, Signature Recognition,
Face recognition, Voice recognition and Iris Recognition form
some of these studies [6].
German language belongs to the Germanic branch of IndoEuropean language family. Approximately 120 million people
are speaking German language in the world. In addition,
Germany has an important standing in respect of economy,
trade, industry and many other fields in the international sense.
For this reason, German language is used in a quite

widespread manner. However, common usage of this language
to such an extent, brings security problems for biometric data
in this field. Accordingly, this calls for the requirement of a
secure, fast automatic voice and speaker recognition.
German language comprises of roots (words) and suffixes
& prefixes and included in the inflected languages if we
consider properties of the German language. German is
written using the Latin alphabet and there are 29 letters in its
alphabet. An article appears before each noun in German.
Words are pronounced as they are written. In addition, it is
distinguished from other languages with various developed
sound shifts and intonation.
Various studies have been made to recognize the sound and
speaker. But, only in a small part of these studies, the gender
recognition study, which is based on sound signal, was able to
be made. The voice recognition systems are reviewed as
independent from the speaker and dependent in two parts. In
case the sound record, which is used in training and test stage,
is the same of the speaker, it means dependent to the speaker,
if not it means independent from the speaker.
Perry and his colleagues have tried to determine the gender
of 10 male 10 female speakers, whose ages varied between 416 by looking at acoustic sound features. They have observed
the formant frequency of two vowels, which dont blend and
basic frequency with the age group differences of the
speakers. The formant frequency has reflected the
characteristic features of the sound and relations between the
age differences and gender differences. But, this study has
been realized on an English data base [7].
Parris and Carey have realized a study, independent from
the text in order to determine the gender of the speaker. They
have tried to develop a new method by combining acoustic
sound features and tone frequency of the speaker. They have
normalized the data with Linear Discriminant Analysis
method LDA- on OGI data base and trained it with Sakl
Markov Model [11].
Various studies have been carried out in order for voice and
speaker recognition. Jie-Fu et al. have collected voice samples
in Chinese from 7 males and 5 females whose ages were
ranging between 25 and 45 [1]. Attempts have been made to
identify the owner of the voice by trying to analyze these
voice samples by means of their tones, vowels, consonants and
syllables.
C. BAKIR, is with Department of Computer Engineering University of Yildiz

Technical University, Istanbul, Turkey, (e-mail: cigdem@ce.yildiz.edu.tr).
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
Voice samples have been separated into four frequency

groups, and each frequency band has been analyzed. However,
this study has not been tested for very big data. In addition,
intended success was not exactly achieved since it was
performed taking its similarities with the English language
into consideration.
Tokuda et al. have developed English speech synthesis
system using Hidden Markov Model [2]. This system has been
developed for speaker recognition and specifies the structure
by changing the voice feature. However characteristic feature
of the synthesized voice in the study, is pretty low.
Reynolds et al. have implemented SuperSID project in order
to enhance performance of speaker recognition systems [3].
Purpose of this project is to develop speaker recognition
systems and employ the most suitable features in order to
increase its accuracy. However, this study failed to completely
achieve the acoustic characteristics of the voice and removal
of the noise.
Reynolds et al. have attempted to substantiate speaker
identification and verification using Gaussian Mixture Model
(GMM) method [8]. Attempts have been made to determine
speaker verification in the system according to the probability
distribution. 11 different hypotheses have been developed for
this probability distribution. Data used in the study has been
extracted from telephone conversations.
Speech has an important place in communication. Voice
recognition study has been carried out for this reason. In this
study, a simulation also has been performed in order to solve
the voice recognition problem related to the security risk.
However, certain difficulties have got in the way while
creating voice database. There was such difficulty ranked first
among the others that words were vocalized at different speeds
and in different pronunciation by different persons. In addition
to that, such reasons as the noise occurred in the environment
and voice while recording the voice data, toning effect and
syllable stress make voice recognition process difficult [9].
Feature extraction and classification techniques used, were
given in the section 2 of the study performed, experimental
study in the section 3 and conclusion were given in the section
4.
II. METHOD
German language is widely used in economy, industry and
trade. Therefore, examinations have been made on German
language in this study. The study has been realized on a
unique data base, which have been formed from the German
sound samples, taken from men and women. These sound
samples are trained by getting dispersed to various feature
vectors with MFCC. In the second stage, the feature vectors of
the recorded sound signals are trained with classification
algorithms, such as DTW, HMM and ANN. The gender of the
speaker is decided by looking at sound signals at the test data
and training data after the system is trained. Furthermore, the
classification success in recognizing the gender of speaker has
been calculated separately for MFCC-1, MFCC-3 and MFCC5 and MFCC-9 and the success of the methods have been
presented comparatively by training the feature vectors,
obtained from speaking signals with DTW, HMM and ANN.
Copyright BAJECE
ISSN: 2147-284X
DOI: 10.17694/bajece.43067
80
A. Mel Frequency Cepstral Coefficients (MFCC)

MFCC is a feature extraction method, that is used in
sound processing. It is used to extract important information
and features by dividing the sound data to its subsets. The
steps of feature extraction technique of MFCC is indicated in
Figure 1[10].
Pre-Emphasis
Framing
Windowing
FFT Spektrum
Mel Spektrum
Mel Cepstrum
Feature Extraction
Fig.1 -MFFC feature extraction steps

Two filters are used in MFCC feature extraction method.
The first filter has a linear distribution of frequency values
under 1000 Hz and the other has a logarithmic distribution of
frequency over 1000 Hz. Pre-emphasis stage is the first stage
in obtaining MFFC feature vector.
The sound signals, which have high frequency, are passed
through a filter at this stage. This way, the energy of the sound
is increased at high frequency. The sound signals are analog.
The sound signals are converted from analog to digital by
getting divided into small frames between 20 and 40 ms
during the framing stage and it is divided into N frames. The
sound signal is moved by sliding the sound signal at the
windowing stage. This way, the closest frequency lines and
the frame, which will come by windowing, that is used are
combined. The window type, width and sliding amount are
determined at this stage. Each of N frames is transmitted from
the time space to the frequency space with Fast Fourier
September 2016
Vol.4 No.2
Transformer (FFT). The spectral features of sound signals are

shown in frequency space.MEL spectrum is obtained by
calculating the total weight of these spectral features. This
MEL spectrum is formed from triangle waves and are formed
by getting passed through a series of filters. MEL spectrum
reduces the noise by lowering two neighbour frequencies. The
logarithm of signal is taken at the stage of MEL spectrum and
the signal is transmitted back again from frequency space to
the time space. MEL frequency cepstrum factors are obtained
by using DCT (Discrete Cosine Transform) in time space.
Copyright BAJECE
ISSN: 2147-284X
81
their weights, to the input of next layers, and these weight are
used in the calculation of outputs for the next layer. Weights
of the hidden layer between input and output layers are
calculated [7].
x1
y1
B. Artifical Neural Network (ANN)

ANNs have a very wide fields of application up to
automotive, banking, defense industry, electronics,
entertainment, finance, insurance, manufacture, oil and gas,
robotics, telecommunication and transportation industry.
Artificial neural networks are information systems which
mirror human brain function, and classify the data through
learning. They have been developed, being based on a
principle of human brain functioning. In other words; ANNs
have been developed with a logic similar to the biological
neural networks, and are data processing structures connected
to each other with weights.
ANNs comprise of input layer, output layer and hidden
layers. Data is received into neural networks through input
layer. And it is transferred to outside through output layer.
Layers between input and output layers constitute hidden
layers.
Neurons in the feed-forward neural networks are connected
just in the forward direction [6]. Each layer of neural network
contains the connection of next layer and these connections
are not in the backward direction. In a sense, there is a
hierarchical structure between neurons, and the neurons
located in one layer can only communicate data to the next
layer. Structure of a feed-forward ANN is shown in the Figure
2.
Backward propagation network shows how to train a neuron
[7]. Trainer is a sort of learning. Network is maintained both
with the sample inputs and expected outputs when the trainer
method is employed. Expected outputs are compared with
actual outputs for the networks the inputs of which are given.
Error is calculated in case the expected outputs are used, and
weights of various layers are adjusted in the backward
direction from output layer to input layer. In other words, it is
given for both input data and output data. Network updates its
coefficients in order to obtain the expected output.
ANN is the most widely used method. In this algorithm,
error in the output layer is calculated at the end of each
iteration, so this error is transmitted to all neurons in the
direction from output layer to input layer, and weights are
readjusted according to the error margin. Such error margin is
distributed to the previous neurons located before the said
neuron in proportion to their weights.
Layers are located one after another in a multilayer artificial
neural network. Outputs of neurons in a layer will be given as
D. Dynamic Time Warping (DTW)
Dynamic Time Warping (DTW) finds out to which
speaker the voice signal given belongs, by calculating the
DOI: 10.17694/bajece.43067
x2
Fig. 2. Feed-forward neural network
C. Hidden Markov Model (HMM)

A lot of studies have been carried out with regard to the
Hidden Markov Models (HMM) in many fields from past to
today. HMM has been used in a wide manner in face
recognition, speech recognition, voice recognition, hand script
recognition, human body motion recognition, bioinformatics,
estimation of gene, cryptanalysis, protein structure and
sequence, DNA sequence and pattern recognition.
In Hidden Markov Model (HMM) the aim is to try to
estimate future situations that will likely occur in cases when
the existing situations are given as an input to the system.
HMM is a stochastic process since it generates different output
whenever it is operated. In addition, system in Markov models
may move from its own state to another state according to the
probability distribution, or remain in the same state.
Probabilities occurred in the states are called as transition
probabilities. States are not seen by the observer as distinct
from HMM normal Markov model. However, transition
subject to the states may be observed. HMM speaker
recognitions systems comprise of the following steps [5].
S= {
,,
} shows current status of the speech
signals generated where there are Q numbers of states.

Initial state probabilities is determined in a discrete
time, t. ( = { ( |t=0, S} )
Transition probabilities are calculated according to
the current states. aij = ( Pr(Sj t in time t| Si in time t-1), Si
S, Sj S))
F, which is the number of features observed, is
determined.
Probability distribution of speech signal will be
calculated in this way. ( bx= {bi (x)= Pr (x(Si ), Si S, x F})
HMM generated is demonstrated by =(a,b, ).
similarity between the time-variant two speech signals. The

most optimal time curve can be identified between two signals
with this method.
September 2016
Vol.4 No.2
Q=
,,
,,
C=
,,
,,
)= (
(2)
)2
82
increased, all techniques, that have been used for success of

recognizing the gender of speaker has also increased. HMM
has given more successful results compared to other
techniques. The tone frequency is between 120-200 Hz in
women and 60-120 Hz in men. The determining the sex in
women from tone frequency is more successful compared to
the men.
(1)
Q and C in the equation 1 and equation 2 demonstrate

two distinct speech signals; n and m show the lengths of these
speech signals [4]. In this case, the ratio of similarity between
Q and C signals is calculated using Euclid length as in the
equation 3 [4].
d(
DOI: 10.17694/bajece.43067
TABLE I
ATTRIBUTES OF USED DATABASES
(3)
Number of speaker
Age Range
A matrix (i,j) is generated for Q and C. Accumulated

distance matrix is calculated using this matrix.
D(i,j)=min[D(i-1,j-1),D(i-1,j),D(i,j-1)]+d(i,j)
(4)
In this study, an authentic and unique German data base has

been used. The names, surnames, ages, sexes and speeches of
persons have been added into this data base. The different
number of feature vectors of sound components have been
extracted with MFCC feature extraction method. In the next
stages, the sound samples have been trained by using methods
of ANN, HMM and DTW. The features of the recorded sound
samples have been indicated in Table 1.
In Table 2, the success ratios of the speech samples,
obtained by taking MFCC-9 Feature Vector, for ANN, HMM
and DTW have been given. As much as the number of the
words, used in speeches, are increased, all techniques, that
have been used for success of recognizing the gender of
speaker has also increased. Sakl Markov Model has given
more successful results compared to other techniques. The
names, surnames, ages, sexes and speeches of persons have
been added into this data base. The different number of feature
vectors of sound components have been extracted with MFCC
feature extraction method. In the next stages, the sound
samples have been trained by using methods of ANN, HMM
and DTW. During the test stage, it has been tried to be defined
whether the test sample is a male or a female with the used
methods. Furthermore, the success of all methods have been
calculated and presented comparatively.
A unique and genuine German language database has been
employed in this study. Names, family names, ages, speeches
and genders of the persons were added to this database.
Feature vectors of voice components with different quantities;
have been extracted by means of MFCC feature extraction
method. Voice samples have been tested by training them,
using available feature vectors by means of ANN, HMM and
DTW methods. In the testing phase it was determined male or
female by available testing example. It has also presented and
compared by calculating the success of any method used.
In Table 2, the success ratios of the speech samples,
obtained by taking MFCC-1, MFCC-3, MFCC-5 and MFCC-9
feature vectors for ANN, HMM and DTW have been given.
As much as the number of the words, used in speeches, are
ISSN: 2147-284X
Female
15
23
12
19
26
15
18-25 range speakers

26-40 range speakers
41 and more speakers
III. EXPERIMENTAL STUDY
Copyright BAJECE
Male
TABLE II
SUCCESS OF THE GENDER SPEAKER USED METHODS
Used
Feature
Vectors
ANN (%)
HMM (%)
DTW (%)
Female
Male
Female
Male
Female
Male
MFCC-1
MFCC-3
MFCC-5
MFCC-9
72.03
74.02
81.63
85.69
68.44
71.87
79.37
80.36
71.63
76.35
84.99
98.34
69.63
75.33
82.37
97.02
69.63
76.01
79.22
87.37
67.35
71.93
75.54
86.33
IV. CONCLUSION
The sound recognition has a great importance from the angle
of security and many other reasons. In this study, a system,
aimed at determining the gender of the speaker, has been
developed on the unique database, obtained by using German
language. Classification success of the methods, used in the
study, has been calculated separately in men and women and
the results have been presented comparatively. When we
looked at the results; it has been seen that HMM method has
given more successful results compared to other classification
methods. Furthermore, the speaker gender recognition system
is more successful in women compared to the men. MFCC-9
feature extraction has rather more success as per the results,
obtained by using 1,3 and 5 feature vectors.
REFERENCES
[1]
[2]
[3]
Quan, Jie-Fu, Fan Gang,

Zeng F and Robert, Shannon etc.,
(Importance of tonal envelope cues in Chinese speech recognition,
The Journal of the Acoustical Societct of America, Vol.104, No.1,
pp.505-510, 1998.
Keiichi, Tokuda , Heiga, Zen and Alan, Black, An HMM- Based
Speech Synthesis System Applied to English, Proc.of 2002 IEEE
SSW, pp.227-230, 2012.
Douglas, Reynold , Walter, Andrews and Joseph, Campbell etc.,The
SuperSID Project: Exploiting High-Level Information for HighAccuracy Speaker Recognition, In.Proc. ICASSP, Hong Kong,
pp.784-787, 2003.
September 2016
Vol.4 No.2
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Lindasalwa, Muda and Mumtaj, Began, Voice Recognition Algorithms

using Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time
Warping (DTW) Techniques, Journal Computing, Vol.2, No.3,
pp.138-143, ISBN 2151-9617, 2010.
Edmondo, Trentin and Marko, Gori, A survey of hybrid ANN/HMM
models for automatic speech recognition, Elsevier Neurocomputing
37, pp.91-126, 2001.
Seok, Oh and Ching, Suen, A class-modular feed forward neural
network for handwriting recognition, Pattern Recognition, vol.35,
issue 1, pp.229-244, 2002.
Theodore L. Perry, Ralph N. Ohde,a) and Daniel H. Ashmead, The
acoustic bases for gender identification from childrens voices, J.
Acoust. Soc. Am. 109 (6), pp.2988-2998, 2001.
Douglas, Reynolds, Thomas, Quatieri and Robert, Dunn, Speaker
Verification using Adapted Gaussian Mixture Models, Digital Signal
Processing 10, pp.19-41, 2000.
Wouter, Gevaert, Georgi, Tsenov and Valeri, Mladenov, Neural
networks used for speech recognition, Journal of Automatic Control,
Vol.20, pp.1-7, 2010.
Lindasalwa Muda, Mumtaj Begam and I. Elamvazuthi, Voice
Recognition Algorithms using Mel Frequency Cepstral Coefficient
(MFCC) and Dynamic Time Warping (DTW) Techniques, Jornal of
Computing, Vol.2, No.3, pp.138-143, ISSN 2151-9617, 2010.
Eluned, Parris, Micheal, Carey, Language Independent Gender
Identification, Acoustics, Speech, and Signal Processing, 1996.
ICASSP-96. Conference Proceedings., 1996 IEEE International
Conference on, Vol.2, pp.685-688, 1996.
Lihang, Li, Dongqing, Chen and Sarang, Lakare etc, Image
segmentation approach to extract colon lmen through colonic material
taggng and hidden markov random field model for virtual
colonoskopy, Medical Imaging, 2002.
Seok, Oh and Ching, Suen, A class-modular feed forward neural
network for handwriting recognition, Pattern Recognition, Vol.35,
No.1, pp.229-244, 2002.
Copyright BAJECE
ISSN: 2147-284X
September 2016
DOI: 10.17694/bajece.43067
83
BIOGRAPHIES
CIGDEM BAKIR was born in stanbul. She
received the B.S. degrees in computer engineering
from the University of Sakarya, in 2010 and the M.S.
degree in computer engineering from Yildiz
Technical University, stanbul, in 2014.
Since 2012, she was a Research Assistant with the
Yildiz Technical University. Her research interests
include recommendation systems, data mining,
image processing and biomedical signal processing.
Vol.4 No.2
DOI: 10.17694/bajece.93236
84
Implementation of Model Reference Adaptive

Controller with Fractional Order Adjustment
Rules for Coaxial Rotor Control Test System
G. Kavuran, B. B. Alagoz, A. Ates, C. Yeroglu
Abstract In this study, an experimental test platform of a

low-cost coaxial rotors has been developed for implementation of
Model Reference Adaptive Controller (MRAC) with Fractional
Order Adjustment Rules by using MATLAB/Simulink. This
paper provides a design method of MRAC with fractional order
adjustment rule (FOAR-MRAC) that is implemented on a
hardware based on ARM microcontroller. Setup of the test
platform is presented and its performance is evaluated with realtime experimental measurements. This experimental study is
useful to show the utilization of fractional order control systems
on engineering applications by using low-cost hardware.
Index Terms Adaptive PID controller, DC rotor control,
fractional order integrator, model reference adaptive control,
TRMS.
I. INTRODUCTION
OAXIAL rotors are composed of twin blade turning in

opposite direction. The idea of coaxial rotors come from
Mikhail Lomonosov who developed a small helicopter model
with coaxial rotors in July 1754 and demonstrated it to the
Russian Academy of Sciences [1]. Turning of twin blades in
opposite direction provides symmetry of forces around the
central axis. This is a main advantage of the system comparing
to single blade rotor systems. This helps solution of problems
of single rotor system such as rotate of main body in the
direction opposite to the rotor blades, dissymmetry of lift.
Main disadvantage of coaxial rotors is the complexity of
mechanical parts.
MRAC was developed by Whitaker around 1960 [2] and
Landau in late 1970 [3]. Later, it turned into a fundamental
topic of adaptive control literature [4, 5]. MRAC was shown
that it is capable of improving robust performance for
parameter variations, noise and uncertain dynamics [6] and
therefore it was widely used in practical applications.
G. KAVURAN, is with Department of Mechatronics Engineering, Firat
University, Elazig, Turkey, (e-mail: gkavuran@firat.edu.tr).
B. B. ALAGOZ, is with Department of Computer Engineering, Inonu
University, Malatya, Turkey, (e-mail: baykant.alagoz@inonu.edu.tr).
A. ATES, is with Department of Computer Engineering, Inonu University,
Malatya, Turkey, (e-mail: abdullah.ates@inonu.edu.tr).
C. YEROGLU, is with Department of Computer Engineering, Inonu
University, Malatya, Turkey, (e-mail: c.yeroglu@inonu.edu.tr).
Copyright BAJECE
ISSN: 2147-284X
For instance, controlling hybrid tank systems with MRAC

approach [7], the vector controlled induction motor control
[8], controlling five-phase interior-permanent-magnet (IPM)
motor [9]. MRAC structure was also utilized for distributed
control applications [10, 11]. Recently, MRAC was
considered for a class of uncertain dynamical systems with
non-linear delayed state perturbations [12].
MRAC mainly used for the control applications where
system dynamics alters by changing the environmental
condition [13, 14]. Flight control requires dealing with
complicated aircraft mechanisms, nonlinear and inaccessible
aerodynamics and frequent variability in flight conditions,
depending on altitude, payload and weather conditions.
Reference model based adaptive control strategies allow autotuning of control systems for a desired system response and
they can improve flight control performance under varying
conditions [15]. Although metaheuristic methods can provide
good solutions for unknown plants in simulations [15], they
are not very effective for real time tuning application due to
the requirement of repetitive set and trail sessions. However,
real-time adaptation skill of MRAC can allow real-time
convergence of feedback control system response to a desired
response of theoretical reference models. It is very useful to
robust performance of control system under varying
conditions of real applications. In general, MRAC method is
very effective for active control systems, of which gain
parameters need to be adjusted when control dynamics
changes. MRAC implements well known MIT rule that
provides the descent of gradient of model approximation error
during update of gain parameters. Essentially, this real-time
self-tuning process leads to adaptation of control system.
During the last two decades, control engineering has been
benefited from fractional calculus in system modeling and
control problems. It was shown in many works that fractional
order controller and system models can considerably improve
control system performance [16-18].
Vinagre et al. modified the conventional MRAC structure
by using fractional order integrator [5]. They theoretically
demonstrated that fractional order adjustment rule MRAC
(FOAR-MRAC) can improve the tracking performance.
In this paper, we developed an experimental test platform for
coaxial rotors control and conduct an experimental adaptive
control study by using for the FOAR-MRAC structure. The
FOAR-MRAC structure is implemented on Arduino card by
using embedding coding technique, and thus adaptive control
operation can be per-formed real-time by low-cost Arduino
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.93236
Mega 2560 card. The FOAR-MRAC structure was designed in

MATLAB/Simulink and then loaded to the Arduino Mega
2560, which control the coaxial rotors test platform. Coaxial
rotors control test platform is composed of a coaxial rotors and
a wooden shaft and it is fixed to container by means of
incremental rotary encoder pin. We tested FOAR-MRAC
structure with different fractional order integrator
configurations in this test platform and discuss results.
control system adapt itself according to a reference model.

The reference model is commonly chosen a theoretical control
system that represents a desired controller performance.
II. METHODOLOGY
A. Foundation of Fractional Calculus
Fractional order derivative operator as an extension of
integer order derivative and integrator were written in general
form as [16],
a Dt
1
f ( )
d , n 1 n
(n ) a (t ) n1
Adaptation
Mechanism
updates the adaptation parameter (t ) by minimizing a

convex cost function J via MIT rule:
(1)
1
J ( ) ed2 ( )
2
(3)
where y r is the reference model output, and y is the

control system output. This error definition enforces the
(4)
The adjustment rule of conventional MRAC for the change

of adaptation parameter in the direction of the negative
gradient of J ( ) was written [3, 5] as,
de
d
dJ

ed d
dt
d
d
(2)
B. Theoretical Background for FOAR-MRAC Structure

Figure 1 illustrates basic blocks of conventional MRAC
proposed for gain adaptive control applications. The
adaptation mechanism of MRAC is based on gradient descent
technique, which was also known as MIT rule [21]. MRAC
method can perform adaptation under the assumption that
system parameters deviate more slowly than the adaptation
parameter of MRAC [5]. The model approximation error can
be defined as,
ISSN: 2147-284X
Whenever responses of reference model and control system

differ, it causes increasing amplitude of ed (t ) and MRAC
The equation (2) leads to very useful property on Laplace

transform of fractional order derivative which was given as
L( D f (t )) s F (s) for zero initial conditions.
Implementation of theoretical fractional order derivative is
not practical because of its high computational complexity.
Integer order approximations of fractional order models are
widely preferred for the practical implementation of fractional
order system models. In this study, we implemented the fourth
order integer order approximate model of fractional order
derivatives by means of CFE method [20].
Copyright BAJECE
yp Plant
Fig. 1. Schematic diagram of MRAC
ed yr y
Controller
where, the operator a Dt denotes non-integer order

derivative of fractional calculus. Parameters a and t are
the lower and upper bounds of the operator, and R
denotes the fractional-order. The Caputo definition of
fractional order differentiation was defined based on (.) ,
namely Eulers gamma function, as follows [19],
t
ym
Reference
Model
d
0
dt
1
0
a Dt
t
(d )( ) 0
85
(5)
Equation (5) provides evolution of adaptation parameter ( )

in the direction reducing the cost function. Later, Vinagre et
al. modified the conventional MRAC structure by using
fractional order integrator and suggested FOAR-MRAC [5] as,
d t
de
dJ
ed d
d
d
(6)
For FOAR-MRAC structure, the adaptation parameter

e
were solved by considering d ym as,
D ed ym
(7)
Then the control signal was scaled by adaptation parameter

as,
u (t ) uc (t )
(8)
where, parameter is for feedforward adaptation parameter

of the control system. The temporal evaluation of
adaptation parameters minimizes cost function according to
fractional order MIT rule [5] given by equation (8),
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.93236
Reference model is assumed to be a second order system in

n2
the form of 2
and the control signal u c (t ) is
s 2n n2
provided by PID controller in closed loop unity feedback
system. Here, FOAR-MRAC provides adaptation by scaling
control effort of closed-loop PID controller as illustrated in
Figure 3. This approach comes out the advantage of easily
transformation of conventional PID control system into
adaptive PID control systems. In our experimental studies, we
used PID coefficients k p 0.6 , ki 0.1 and kd 0.3 .
86
and blades set of LS-222 Gyro 3.5 Channel model helicopter

as shown in Figure 4.
Fig. 4. Close views of coaxial rotors and blades used in the experimental
system.
+
-
PID
uc
usat
yp
Fig. 2. Block diagram of the proposed FOAR-MRAC structure
III. EXPERIMENTAL STUDY

A. Coaxial Rotors Control Experimental Test Platform
Mechanical parts of coaxial rotors control experimental test
platform is similar to well-known TRMS system produced by
Feedback Instruments Inc. Coaxial rotors control experimental
test platform composed of a coaxial rotor, a wooden shaft,
incremental rotary encoder (ES5-0CCN 6942) and carton
container. The shaft is made of light and flexible model
woods that reduce transmission of high frequency vibration of
coaxial rotors to rotary encoder. The carton container fixes
rotary encoder holding the wooden shaft. Figure 3 depicts a
prototype of experimental system.
For the control of test platform, we used Arduino Mega 2560

card. Arduino platforms are frequently used for experimental
unmanned aerial vehicle (UAV) studies. The FOAR-MRAC
structure was designed in MATLAB/Simulink and the design
was loaded to the Arduino control card. This card receives the
angle data of shaft from rotary encoder and drives two
electrical motor of coaxial rotors by means of Darlington
power amplification circuit. Figure 5 shows circuit diagram of
electronic parts of the experimental system. Advantages of
implementation of FOAR-MRAC by embedded programming
on Arduino Mega 2560 card that it is a low-cost control card
and it supports embedded programming with Matlab Simulink
environment. MATLAB/Simulink design for FOAR-MRAC
structure is illustrated in Figure 6.
5V
5V
2K
2K
12V
2K
(A)
V out2
(B)
V out2
Fro m Enco der (B)
V in2
Fro m Enco der (A)
ENCODER
V out1
V out1
V in1
(C)
Dri ve r
Fig. 5. Electronic circuit scheme of experimental system.
Fig. 3. A prototype of coaxial rotors controls experimental test platform
Coaxial rotors consist of twin blade turning in opposite

direction. Turning of twin blades in opposite direction
provides symmetry of forces around the central axis. This is
useful for dealing with some complications of single rotor
system such as rotate of main body in the direction opposite to
the rotor blades, dissymmetry of lift. We used coaxial rotors
Copyright BAJECE
ISSN: 2147-284X
Fig. 6. MATLAB/Simulink design for FOAR-MRAC structure
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.93236
1.6
1.4
1.2
0.8
10
20
30
40
50
60
70
t (sec)
Fig. 9. Change of adaptation parameter (t ) for FOAR-MRAC structure
with the fractional order integration of 0.8
2.2
2
1.8
1.6
(t ) stabilizes in time. Because, it converges a local minima

of the objective function J ( ) and thus the adaptation process
of FOAR-MRAC takes places.
We observed that step responses are very smooth and
ripples are very low. One of reasons is the materials used in
test platform, which are light and flexible woods and carton
container. They can reduce the transmission of mechanical
vibration caused from blades and rotor motion.
1.8
B. Experimental Results
In this section, test results obtained for FOAR-MRAC
implementations are presented for the fractional order
integration of 0.88 and 1.12. Figures 7 and 8 show the step
responses of FOAR-MRAC implementations. For the
fractional order integration of 0.88, it settles 0.7 radian
without any overshooting in Figure 7 and this smooth settling
is a desired response for flight control [15]. Figure 8 shows
integration with the order of 1.12 provides faster rise but it
gives slight overshoots and this response may result in
disturbance for multi-rotors systems such as quadcopter.
Because, all rotors are coupled with each other and overshoots
in settling may cause swing of multi-rotor system.
Figures 9 and 10 show the change of adaptation parameter
(t ) during step responses. One can see that the change of
87
1.4
1.2
0.8
0.7
0.8
0.6
(rad)
p
0.4
10
20
30
40
50
60
70
t (sec)
Fig. 10. Change of adaptation parameter (t ) for FOAR-MRAC structure
with the fractional order integration of 1.12
0.5
0.3
IV. CONCLUSION
0.2
0.1
0
10
20
30
40
t (sec)
50
60
70
Fig. 7. Step response of FOAR-MRAC structure with the fractional order

integration of 0.88
0.8
(rad)
0.6
0.4
0.2
Rotor control application requires adaptive control

techniques due to the changing dynamics of flight conditions.
This study demonstrated an experimental study for
implementation of FOAR-MRAC method for the propose of
adaptive rotor control of coaxial rotors experimental test
platform. The FOAR-MRAC structure suggested by Vinagre
et al. was implemented on Arduino card. Indeed, FOARMRAC method was benefited to modify closed loop PID
control system and thus turns the conventional PID control
system into an adaptive PID control system with FOARMRAC support. Moreover, the coaxial rotors control
experimental test platform developed was introduced in detail
and experimental results were discussed briefly. In future
study, we plan to add gyroscope sensors to the test platform
and conduct experiments for adaptive coaxial rotors control of
FOAR-MRAC structure by providing gyroscope sensors
feedbacks.
ACKNOWLEDGMENT
10
20
30
40
t (sec)
50
60
70
This research is supported by TUBITAK with the project id:

215E261.
Fig. 8. Step response of FOAR-MRAC structure with the fractional order

integration of 1.12
Copyright BAJECE
ISSN: 2147-284X
September 2016
Vol.4 No.2
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
ISSN: 2147-284X
88
BIOGRAPHIES
J.G. Leishman, Principles of Helicopter Aerodynamics, Cambridge

University Press, p. 8, 2006.
P.V. Osburn, H.P. Whitaker, A. Kezer, Comparative studies of systems,
Institute of Aeronautical Science, pp. 6139, 1961.
I.D. Landau, Adaptive Control the Model Reference Approach, Marcel
Dekker: NY, USA, 1979.
K.J. Astrom, B. Wittenmark, Adaptive Control, Addison-Wesley,
Reading, MA, USA, 1995.
B.M. Vinagre, I. Petrs, I. Podlubny, Y.Q. Chen, Using fractional order
adjustment rules and fractional order reference models in modelreference adaptive control, Nonlinear Dynamics, Vol.29, pp. 269-279.
2001.
M.D. Bernardo, U. Montanaro, J.M. Olm, S. Santini, Model reference
adaptive control of discrete time piecewise linear systems, International
Journal of Robust and Nonlinear Control, Vol.23, pp.709-730. 2013.
K.A. Mohideen, G. Saravanakumar, K. Valarmathi, D. Devaraj, T.K.
Radhakrishnan, Real-coded Genetic Algorithm for system
identification and tuning of a modified Model Reference Adaptive
Controller for a hybrid tank system, Applied Mathematical Modeling,
Vol.37, pp.3829-3847, 2013.
T.A.V. Ravi, C. Chakraborty, S. Maiti, Y. Hori, A new model reference
adaptive controller for four quadrant vector controlled induction motor
drives, IEEE Transactions on Industrial Electronics, Vol.59, pp.37573767, 2012.
L. Guo, L. Parsa, Model Reference Adaptive Control of Five-P IPM
Motors Based on Neural Network, IEEE Transactions on Industrial
Electronics, Vol.59, pp.1500 1508, 2012.
Y. Liu, Y. Jia, Adaptive leader-following consensus control of multiagent systems using model reference adaptive control approach, IET
Control Theory & Applications, Vol.6, pp.2002-2008, 2012.
Z. Peng, D. Wang, H. Zhang, G. Sun, H. Wang, Distributed model
reference adaptive control for cooperative tracking of uncertain
dynamical multi-agent systems, IET Control Theory & Applications,
Vol.7, pp.1079-1087, 2013.
H. Wu, M. Deng, Robust adaptive control scheme for uncertain nonlinear model reference adaptive control systems with time-varying
delays, IET Control Theory & Applications, Vol.9, pp.1181-1189,
2015.
B. Jiang, Z. Gao, P. Shi, Y. Xu, Adaptive fault-tolerant tracking control
of near-space vehicle using TakagiSugeno fuzzy models, IEEE
Transactions on Fuzzy Systems, Vol.18, pp.1000-1007, 2010.
I. Sadeghzadeh, A. Mehta, Y. Zhang, C.A. Rabbath, Fault-tolerant
trajectory tracking control of a quadrotor helicopter using gainscheduled PID and model reference adaptive control, Annual
Conference of the Prognostics and Health Management Society,
Montreal, Canada, pp. 1-10, 2011.
B.B. Alagoz, A. Ates, C. Yeroglu, Auto-tuning of PID controller
according to fractional-order reference model approximation for DC
rotor control, Mechatronics, Vol.23, pp.789797, 2013.
I. Petras, Fractional-order nonlinear systems: modeling, analysis and
simulation, Springer, Higher Education Press, Beijing, China, 2011.
C. Yeroglu, A. Ates, A stochastic multi-parameters divergence method
for online auto-tuning of fractional order PID controllers, Journal of the
Franklin Institute, Vol.351, pp.2411-2429, 2014.
C. Yeroglu, G. Kavuran, Sliding mode controller design with fractional
order differentiation: applications for unstable time delay systems,
Turkish Journal of Electrical Engineering & Computer Sciences, Vol.22,
pp.1270-1286, 2014.
D. Sierociuk, I. Podlubny, I. Petras, Experimental evidence of variableorder behavior of ladders and nested ladders, Control Systems
Technology, Vol.21, pp.459-466, 2013.
Y.Q. Chen, B.M. Vinagre, I. Podlubny, Continued fraction expansion
approaches to discretizing fractional order derivatives an expository
review, Nonlinear Dynamics, Vol.38, pp.155-170, 2004.
E. Lavretsky, Adaptive control: Introduction, overview, and
applications, NASA Adaptive Control Workshop 2009, NASA Marshall
Space Center, Huntsvill, USA, 2009.
Copyright BAJECE
DOI: 10.17694/bajece.93236
GURKAN KAVURAN received his B.Sc. degree in

Electrical and Electronics Engineering from Firat
University in 2008. He received his master degree in
Electrical and Electronics Engineering from Firat
University in 2011. From 2008 to 2010, he was a
Research Assistant with the Electrical and
Electronics Engineering Department, Bozok
University. Since 2010, he has been a Research
Assistant with the Mechatronics Engineering
Department, Firat University. His research interests
include robotics, fractional calculus, control theory
and applications, modeling and simulation, signal processing.
September 2016
BARIS BAYKANT ALAGOZ received bachelor

degree in Istanbul Technical University, Department
of Electronics and Communication Engineering in
1998, M.Sc. and Ph.D. degrees in Inonu University,
Department of Electrical-Electronics Engineering in
2011 and 2015. His research interests include
modeling and simulation of physical systems, control
systems, smart grid. He is working at Computer
Engineering Department in Inonu University.
ABDULLAH ATES received his B. Sc. Degree in

Electrical and Electronics Engineering from Erciyes
University in 2011. He received his master degree in
Computer Engineering from Inonu University in
2013. His research interests include optimization
algorithm, analytical and numerical optimization
method, fractional order control systems, fractional
order calculus and its application, linear and
nonlinear optimization algorithm flight control.
CELALEDDIN YEROGLU received his B.Sc.

degree in Electrical and Electronics Engineering
from Hacettepe University in 1990. He received his
first Ph.D. degree in Computer Engineering from
Trakya University in 2000 and his second Ph.D.
degree in Electrical and Electronics Engineering
from Inonu University in 2011. His research
interests include fractional order control systems,
robust control, nonlinear control, modeling and
simulation, optimization algorithm, flight control.
Vol.4 No.2
DOI: 10.17694/bajece.39364
89
Determination of Leakage Reactance in

Monophase Transformers Using by Cascaded
Neural Network
H. S. Nogay
AbstractIn this study, the artificial neural networks method

was used in order to determine the leakage induction coefficient
using known constants of the transformer. It is important to
reach the leakage reactance through guessing. For this purpose,
two types of ANN models were used in the study and were
compared to one another. These two ANN models are cascaded
ANN model and the conventional model. Testing data were used
to measure the efficiency of these two models. Testing data are
the same for both models. When the models were compared to
each other, it was concluded that Cascaded ANN model was more
successful. However, it is a fact that both models produce
estimation around 99%. The main reason why the ANN model is
used in the study is to ensure a more practical and quicker
attainment of leakage reactance or leakage induction coefficient
by looking at the fixed and measurable values of the transformer
than calculation method.
Index Terms Transformator, Leakage Induction Coefficient,
Cacaded Neural Network, Backpropagation Learning Algorithm.
I. INTRODUCTION
Investigating of transformers which have a great deal of
importance for production, transmission and distribution of
electrical energy begins with inventing flux distribution. In
this distribution various fluxes can be distinguished from each
other. Leakage flux rung with only one winding plays an
important role in the theory of electrical machines and
transformers. Determination of the leakage field distribution in
transformers is required to calculate inductances, the force
applied to the windings, additional losses occuring in the
windings and iron. Icreasing energy consumption and the
growing generator power required large power transformers.
Despite the restrictions of materials, dimensions and weight,
performing suitable transformers production has been made
possible with advances of leakage flux determination [1, 2].
Leakage flux generated by primary and secondary windings
currents give rise to internal voltage in primary and secondary
windings. Hence leakage fluxes decrease the useful fluxes. As
a result voltage reduction is seen in output of secondery
windings. Voltages created by leakage fluxes is full inductive
and 90 degree ahead from current.
H.S. NOGAY, works in University of Erciyes, Kayseri, Turkey, (e-mail:

hidirselcuknogay@gmail.com , nogay@erciyes.edu.tr ).
Copyright BAJECE
ISSN: 2147-284X
Hence fluxes leakages are showed as reactances which

connected to transformer circuit as serial and named leakage
reactance[2,3].In this study, the leakage induction coefficient
in mono-phase transformers was estimated using two different
artificial neural network estimation model. It is possible to
find out the leakage reactance value that is supposed to be
found through calculation in order to design a transformer of
any type or power capacity by using the numerical values of
transformer coil width, fictive iron frame height, coil height,
type of coil used and the thickness of no conducting layer. It
has a great deal of importance for designers to determine the
leakage currents in transformers. Leakage induction
coefficient can be calculated based on the different winding
types and core types of the transformer. In addition, it is of
great importance that a designer has an opinion on the leakage
reactance of all transformers without delay. Software or
prediction software that will allow the designer to save time
and will apply to all types of mono-phase transformers is an
essential need for designers and producers. There are a
number of estimation and prediction software and as many
estimation methods on the market. The most commonly used
estimation method is the artificial neural networks. Of many
ANN prediction models, the most efficient and the most
common model is the multi-layer back propagation ANN
model. In this study, the multi-layer back propagation ANN
model is defined as the conventional ANN model. The ANN
model recommended in line with the purpose of the study the
Cascade ANN prediction model which guarantees prediction
with a higher accuracy level compared to the conventional
ANN prediction model. In this study, both methods were used
and the results were compared. The most important feature of
the study is that, both the conventional ANN and cascade
ANN ultimately form a basis for a generalized design
program. In other words, both ANN models are of types that
may easily serve the designer in the design of a mono-phase
transformer of any type. Based on the results of the study,
there is an significant difference between the cascade ANN
prediction model and the conventional ANN prediction model
in favor of the cascade model [4-6].
The architecture of the study is as follows:
In the second part of the study, a mathematical expression
of leakage reactance was given for all coil types in monophase transformers.
September 2016
Vol.4 No.2
DOI: 10.17694/bajece.39364
90
In the third part of the study, basic information was given

on artificial neural network models and the selection of ANN
model architectures used in the study, formation of the data set
and performance of ANN model testing was explained.
In the fourth part of the study, the results attained from the
study were evaluated and discussed.
In the fifth part, which is the final part is the conclusion.
The result of the study is explained in plain terms.
one of which is on the left and the other is the right side of the
Y-Y' axis provides a conclusion by adding up all the results
obtained by considering these magnetic circuits separately in
calculations.
So, considering that both parts divided into two with the Y
Y axis is not different from the simple cylindrical coil form
that we previously examined, the leakage induction coefficient
of the secondary winding can be expressed as follows [8].
II. MATHEMATICAL EXPRESSION OF LEAKAGE REACTANCE IN

TRANSFORMERS
N2
4 2 2 2 r
a a2
k R 1
h
3
9
10 H
(4)
A. SIMPLE CYLINDRICAL COIL FORM

Leakage flux of most common medium centered cylindrical
coil transformers under normal operation conditions and short
circuits is shown in Figure 1. Leakage self-induction
coefficients of primery and secondery windings is as follow .
l1 4.1 2 10 e 0 .10 9
a2
2a1
a2
(1)
l 2 4. 2 2 20 e 0 .10 9 H
(2)
2
2
Leakage induction coefficient (N2) reduced to secondary

winding can be expressed as follow [7, 8]:
hs
N2
8 2 2 2 r
3
k R
h
a
1 a2
9
10 H
(3)
r
a1
1
a2
Fig. 2. Leakage field for two -layered cylindrical coil form

hs
r2
C. SYMMETRIC SLICE, MIXED COIL FORM
r1
Z
Z
Fig. 1. Leakage field for simple cylindrical coil form
B. TWO LAYERED CYLINDRICAL COIL FORM

In the common medium centered winding as shown in
Figure 2, there are two equal nonconductive layer between
them and the fact that the section shown in the figure is
symmetrical to Y-Y axis in terms of form and leakage area
and it can be divided to two magnetic circuits which are not
connected to each other in terms of leakage magnetic circuit
Copyright BAJECE
ISSN: 2147-284X
As can be seen in Figure 3, if this type of winding more

symmetrical than the horizontal X-X axis in the form of
magnetic circuit form which has not connection to it as many
as the number of its nonconductive layers or the section form
of which intersects this in terms of magnetic area and form,
then it can be considered as the two-layered cylindrical
winding section made side-by-side as m which is the number
of high voltage coils (90o rotated). Also, m number of high
voltage coils and low voltage coils are connected in series
among themselves, considering the number of windings of one
of these section parts, the formula for the two-layer cylindrical
winding transformer (4) is written down, multiplied with m
and the leakage induction coefficient can be expressed as
follows [8].
N2
4 2 2 2 r
a a2
k R 1
hm
3
September 2016
Vol.4 No.2
9
10 H
(5)

Z
hs
2 /4
1 /2
2 /2
1 /2
2 /4
hs
Z
Fig. 3. Leakage field for symmetric slice, mixed coil form
D. ZIG-ZAG COUPLING TWO LAYERED CYLINDRICAL COIL

FORM
Since the low voltage winding is to be zigzag-connected,
this winding consists of two parts in the form of two interlaced
cylinders as shown in Figure 4 and the currents on these parts
are currents of two different phases. That is why, with the
consideration that it would be simpler to express the N1 selfinduction coefficient translated to the primary winding as the
leakage self-induction coefficient to be calculated in the case
of this type of winding, the expression of this coefficient is
proven to be made in a similar way to the one that is followed
in the case of simple cylindrical coil [8 - 11].
N1
8 2 1 2 r
a a 2 a 2 9
k R 1

10 H
h
3
3 36
(6)
Z
a1
a2 2
a
2 2
hs
h
DOI: 10.17694/bajece.39364
91
having a memory of its own. In other words, artificial neural

networks are computer programs that imitate the neural
networks.
Artificial neural networks are self-learning mechanism
which does not require conventional abilities from a
programmer.
There are a number of artificial neural networks and the use
of some is more common than that of others. The most
common artificial neural network model is Back propagation
ANN.
Two types of ANN were used in this study. Both ANN
models are of back propagation ANNs. The ANN model
consisting of a single model was named the conventional
model in this study. The model connected to one of the
outputs of four ANN models was named "Cascaded ANN
model"
Cascaded ANN model has no difference from the
conventional model in terms of model type. The only
difference is that it consists of the multiple ANN models. The
more the number of input parameters affecting the result is in
artificial neural networks, the quicker and more accurate the
model learns. Therefore, it is of great importance that the
number of input variables is high in ANN models. And the
input number used in the only ANN model used as the
conventional model in this study is 6. In cascaded model, 10
input numbers were used in total, since each ANN model that
is used would produce one input number. The most significant
advantage of the cascaded model over the conventional model
is that it may have more inputs [12].
Type of models used for this paper is multi-layered backfeed ANN models. These ANN models are used in many of
the studies based on prediction. The multi-layered artificial
neural networks learn from back propagation learning
algorithm. As learning algorithm, Levenberg Marquardt
learning algorithm, one of the back propagation learning
algorithms, has been used [13].
The back-propagation learning algorithm is presented below
in brief. For each neuron in the input layer, the neuron outputs
are given by
ni oi
(7)
where n i is the input of neuron i, and o i the output of

neuron i. Again for each neuron in the output layer, the neuron
inputs are given by
Z
Fig. 4. Equivalent Leakage Magnetic Circuit for Zig-Zag Coupling

Transformer
III. DEVELOPMENT OF THE ANN FORECASTING MODELS

A. ARTIFICIAL NEURAL NETWORK (ANN)
Artificial neural networks are parallel and distributed data
processing structures developed with inspiration from the
human brain, connected to each other with weighting
connections and consisting of processing components each
Copyright BAJECE
ISSN: 2147-284X
nk
Nj
w kj o j
j1
k 1,2,3,..., N k
(8)
where wkj, is the connection weight between neuron j and

neuron k, and Nj, Nk the number of neurons in the hidden layer
and output layer, respectively. The neuron outputs are given
by
September 2016
Vol.4 No.2
ok
1
f k n k , k
1 exp n k k
(9)
where k is the threshold of neuron k, and the activation

function fk is a sigmoidal function. For the neurons in the
hidden layer, the inputs and the outputs are given by
relationships similar to those given in Eqs. (8) and (9),
respectively.
The connection weights of the feed-forward network are
derived from the inputoutput patterns in the training set by
the application of generalized delta rule. The algorithm is
based on minimization of the error function on each pattern p
by the use of steepest descent method. The sum of squared
errors Ep which is the error function for each pattern is given
by
EP
1 k
t pk o pk
2 k 1
Np
(11)
p 1
where Np is the number of inputoutput patterns in the

training set. When an input pattern p with the target output
vector tp is presented, the connection weights are updated by
using the following equations:
w kj pk o pj w kj p 1
(12)
N 2 t
0.1 N 2 max N 2 (t ) 0.9 N 2 min N 2 (t )

N 2 max N 2 min
(16)
Where : N2 and N 2 are the Leakage Induced coefficient of

conventional ANN model time series non-normalized and
normalized, respectively; Nmax and Nmin are the maximum
and the minimum absolute value of the Leakage Induced
coefficient respectively. This normalization function was used
for the other ANN models in this study. [17]
In this study, a single data set was used for two types of
ANN models. In the first YSA model called the conventional
ANN, 6 inputs or, in other words the [6x341] matrix and 1
output of [1x341] was used. Inputs and output used in the
conventional model can be seen in Table 1.
TABLE I. INPUTS AND OUTPUTS OF CONVENTIONAL ANN MODEL
Inputs
Output
R
h
N2
Coil height (mm)
hs
Fictive iron frame height (mm)
a1
Width of the first winding (mm)
a2
Width of the second winding (mm)
Cf
Leakage
Induced
coefficient
Coil form
Insulating layer (mm)
(13)
where is the learning rate, and is the momentum constant.

Again, the connection weights between input layer neuron i
and hidden layer neuron j can be updated by using the
following equations:
w ji pk o pj w ji p 1
(14)
pj o pj 1 o pj pk wkj
Nk
(15)
k 1
It is important to note that the threshold of each neuron is

learned in the way same as that for the other weights. The
threshold of a neuron is regarded as a modifiable connection
weight between that neuron and a fictitious neuron in the
previous layer which always has an output value of unity [14 16].
In order to use the ANN simulator for any application, first
the number of neurons in the layers, type of activation
function (purelin, tansig, logsig), the number of patterns, and
the training rate must be chosen.
Copyright BAJECE
ANN designing process involves five steps. These are

gathering input data, normalizing the data, selecting the ANN
architecture, training the network, and validation-testing the
network. In this paper, the data were normalized according to
(16), also these values ranged from 0.1 to 0.9 for all ANN
models.
B. CREATION OF DATASET
Ep
pk t pk o pk o pk 1 o pk
92
(10)
where tpk is the target output for output neuron k, and opk the
calculated output for output neuron k. The overall measure of
the error for all the inputoutput patterns is given by
DOI: 10.17694/bajece.39364
ISSN: 2147-284X
Input matrix used for the conventional model forms the

input of the first ANN in the first layer of the Cascaded ANN
model. In other words, the input matrix used for both ANN
systems are the same. Cascaded model consists of four ANN
models connected to each other and outputs to inputs. The first
ANN model in the Cascaded ANN model has two outputs.
Inputs and outputs of four ANN models in the Cascaded
ANN system is given in Table 2 below [17, 18].
TABLE II. INPUTS AND OUTPUS OF CASCADED ANN MODEL
Models
Inputs
Outputs
YSA1
R: Inputs of Conventional
Model
YSA2
R1 : 1, 2 , R
YSA3
R2 : R1, r
YSA4
R3 : R2, kR
: Number of turns
of the first winging
Number of turns
of the second winging
r : The average radius
of the windings
kR : Rogowski
Coefficient
N2 : Leakage Induced
coefficient
Structural values of transformers used in the data set being

September 2016
Vol.4 No.2
s, h, 2, 1, , Cf, r ,a1 and a2 were formed according to the

restrictions in Table 3. Data set was formed based on the
transformer sizes that can be designed to form a data set,
increasing the section of the conductor with 0.2, and
calculating the sizes corresponding to the increase on the
sections of each conductor. The wider the range of data set is
kept, the more the transformer sizes which the study addresses
[18, 19].
TABLE III. SUMMARY OF THE DATA SET USED FOR ANN MODELS
Abbreviation
Max.
Min.
s
3
2
h
1000
200
2
5328
100
a2
48
2
a1
96
4
1
2656
50
a1+a2+
146
8
k
12.5
3.424658
kR
0.987261
0.953503
2
2
Cf
4
1
hs
1048.764
202.5806
r
222
141
N2
6.124687
0.001372
DOI: 10.17694/bajece.39364
The input and output patterns used in the cascaded system

for the first step ANN model that named ANN 1 are arranged
with the same inputs of conventional ANN model and two
outputs that represent the forecasting 1 and 2. Inputs and
outputs for the other steps of cascaded system are illustrated as
shown in Fig. 6 [22].
ANN 1
Input
Layer
ANN 2
Hidd en
Layer
a1
Input
a2
Hid de n
Layer
Output
a1
a2
N2
hs
cf
Bias
Fig. 5. Tradional ANN Model
Copyright BAJECE
ISSN: 2147-284X
Output
hs
R1
cf
Bias
Bias
ANN 4
ANN 3
Inp ut
Layer
Hidd en
Layer
Output Layer
C. SELECTING THE ANN ARCHITECTURE

The input and output patterns used in the first model that is
conventional model are arranged with 6 inputs and one output
as shown in Fig. 5. The best number of neurons in the hidden
layer that better adapted to the dataset was selected among
different architectures for all model in the study. The
architecture with 20 hidden nodes presented smaller error on
the validation set during the trainings. Therefore, the
architecture of the selected ANN was 6 -20 -1 for the
conventional model. All ANN models used in this study were
developed using Multi-Layer Perceptron network with three
layers: input layer; hidden layer, and the output layer.
Hyperbolic tangent sigmoid function for all of the layers was
used. The architecture of the proposed cascaded system was
created using by four ANN models [20, 21].
93
Hidd en
Layer
Input
Layer
Output
r
R2
kr
Hidd en
Layer
Input
Layer
Output
R3
N2
R1
R2
Bias
Bias
Fig. 6. Proposed Cascaded Neural Network
D. TRAINING THE ANN MODEL

Neural Networks learn from examples and are generated to
be able to generalize the acquired knowledge during the
training. A suitable strategy for training the neural networks
can affect substantially their generalization ability. In this
study all ANN models were trained with the back propagation
(Levenberg Marquardt) training algorithm. In the training
process of this study, the actual outputs of ANN models were
compared with the desired outputs [22].
In the conventional model; the training set consists of six
input and one output data. The number of data was 341. 70%
of this data (239) were used for training.
In the Cascaded system; the training set consists of six input
data that same as the conventional models inputs and two
output data for the first step ANN 1. For the second step ANN
2, the training set consists of eight input and one output data,
for the third step ANN 3 , the training set consists of nine
input and one output, finally for the forth step ANN 4 the
training set consists of ten input and one output data. The
number of data was same to conventional model in the all
September 2016
Vol.4 No.2
Output-Tradional
Output-Cascaded
16
31
94
Target
Leakage Reactance Coefficient
cascaded steps [22, 23].

The network adjusted the weighting coefficients that began
with random set. The training process has been stopped when
the error has become stable. Training process of the
conventional ANN model and cascaded system are shown in
Fig. 7 and Fig. 8 respectively. ANN simulator has been trained
through the 37 epochs in the conventional system and 84
epochs in the final step of the cascaded system, as shown in
Figure 7 and Fig. 8 respectively.
DOI: 10.17694/bajece.39364
5
4
3
2
1
0
-1
46
Testing Data Pattern Value
Fig. 9. Comparison of the target and ANN results for testing
Fig. 7. Training process of the conventional ANN model

Fig. 10. Regression results of conventional ANN model for testing
Fig. 8. Training process of the cascaded system

Fig.11 Regression results of cascaded ANN model for testing
E. TESTING THE ANN MODEL

At the test stage, 51 data selected as a random set from the
entire dataset were used for all ANN models in this study as
shown in Table 4. In order to be positive about the test result,
validation data were used at 51 pieces. According to the result
of test conducted, as can be inferred from Table 4, a prediction
rate of 97.7% was achieved in the conventional model, and
99.9% was achieved in the proposed system. In Fig. 9
comparison of the target and ANN results for testing is
illustrated. In Fig. 10 and 11, the regression curve obtained
from testing is presented for conventional model and cascaded
model respectively [23, 24].
Copyright BAJECE
ISSN: 2147-284X
TABLE IV. RESULTS OF THE ANN SYSTEMS

Samples
MSE
Conventional
Training
239
0.00352
ANN Model
Validation
51
0.0119
Testing
51
0.0125
Cascaded
Training
239
3.0594x10-6
ANN Model
Validation
51
5.210x10-5
Testing
51
1.180x10-5
R
0.999454
0.997875
0.977540
0.999999
0.999994
0.999997
IV. RESULTS AND DISCUSSIONS

For a better evaluation of the study, 51 data used for the test
were divided into five parts and five different charts were
obtained. Figure 12 shows the comparison with the first ten of
September 2016
Vol.4 No.2
the testing data.
DOI: 10.17694/bajece.39364
produces a more satisfactory prediction compared to the

conventional system.
Output-Tradional
Output-Cascaded
Target
Output-Tradional
0,7
Output-Cascaded
Target
6
Leakage Reactange Coefficient
Leakage Reactange Coefficent
95
0,6
0,5
0,4
0,3
0,2
0,1
5
4
3
2
1
10
10
11
When we compare the results of the conventional system

from the first parts of test results in Figure 12 with the target,
it can be said that two charts are not consistent, though pretty
much close to each other. In addition, when the cascaded
system and the target is compared, the green-colored chart
indicating the output of cascaded system is under the redcolored chart in Figure 12 and can not be seen. In other words,
according to this first party of test results, the results of the
cascaded system and the target are consistent. Based on this
chart, the cascaded system can be said to be more successful
in comparison with the conventional system for the first part.
If we examined Figure 13 obtained from the second part of
test results; it can be seen that the results of the cascaded
system are consistent with the target. On the other hand, the
results of the conventional ANN model are not consistent with
the target, though quite close to it. If we compare the first part
of the test with the second, it can be said that the results of the
conventional system in the second part seem to be more
successful.
Output-Cascaded
Fig. 14. Comparison of the ANN outputs and target for the thirth part of
testing data
If we examine the charts obtained from the fourth and fifth

party test results as shown in Figure 15 and Figure 16, it can
be easily said that the cascaded model is completely consistent
with the target and the conventional model produces
predictions pretty close to the target.
Output-Tradional
Output-Cascaded
Target
4,5
Fig. 12. Comparison of the ANN outputs and target for the first part of testing
data
Output-Tradional
Testing Data (22-32)
Testing Data (0-10)
4
3,5
3
2,5
2
1,5
1
0,5
0
-0,5
10
11
Fig. 15 Comparison of the ANN outputs and target for the forth part of testing
data
Target
Output-Tradional
Output-Cascaded
Target
4,5
4
4,5
3,5
3
2,5
2
1,5
1
0,5
0
-0,5
10
11
4
3,5
3
2,5
2
1,5
1
0,5
0
-0,5
Fig. 13. Comparison of the ANN outputs and target for the second part of
testing data
If we examine the chart in Figure 14 created from the third

part of test results; it can be observed that the cascaded system
indicated with green color in the data between 2-4 of the chart
is distant from target, and is consistent with the target on other
data spots. The results of the conventional system is not
completely consistent with the target but pretty much close. In
this chart, it can be understood that the cascaded system
Copyright BAJECE
ISSN: 2147-284X
Fig. 16 Comparison of the ANN outputs and target for the fifth part of testing
data.
V. CONCLUSION
In this study, leakage inductions (N2) in different coil forms
of mono-phase transformers were predicted by using the
artificial neural networks system. It is obvious that this study
will lay the groundwork for the production of software that
September 2016
Vol.4 No.2
can be used in order to design mono-phased transformers. In

this study, the cascaded model with can be attained by
connecting the ordinary models end to end, instead of the
ordinary ANN prediction model which is always used for
prediction. Because, though the number of inputs is the same
in the cascaded model, which means that ANN 1 has the same
number of inputs with the conventional model, the result is in
favor of the cascaded model. So, the prediction results
generated in the cascaded model have proven more accurate
than the conventional model. This forecasting model applies
for all mono - phase transformers and is a very successful
model in accelerating the designers working in this field and
making their jobs easier through provision of preliminary
ideas on future studies.
VI. LIST OF SYMBOLS AND ABBREVIATIONS
N2 : Leakage Induced coefficient
: Insulating layer
1 : Number of turns of the first winding
2 : Number of turns of the second winding
h
: Coil height
hs : Fictive iron frame height
kR : Rogowski Coefficient
a1 : Width of the first winding
a2
: Width of the second winding
Nj, Nk : The number of neurons in the hidden layer and
output layer
wkj
: Connection weight between neuron j and
neuron k
ni
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
: Input of neuron I
oi
: Output of neuron I
k : Threshold of neuron k
fk : Activation function
Ep : Sum of squared errors
: Momentum constant
s : Cable section
r : The average radius of the windings
ANN : Artificial Neural Network
cf : Coil form
REFERENCES
[1]
[2]
[3]
[4]
[5]
L. Andriuien, A. Kairys, P. Kostrauskas, Mathematical Modeling of

External Characteristics of Power Transformers , Electronics and
Electrical Engineering. Kaunas: Technologija, No. 5, 40, 2002
P. Raitsios, Leakage field of a transformer under conventional and
superconducting condition, Journal of Materials Processing
Technology, 108. 246-252. 2001.
R. Doebbelin, C. Teichert, M. Benecke, and A. Lindemann,
Computerized Calculation of Leakage Inductance Values of
Transformers, Piers Online, Vol. 5, No. 8, 2009.
M. A. Tsili, A. G. Kladas, S. Georgilakis, Computer aided analysis and
design of power transformers, Computers in Industry, 59, pp. 338350,
2008
A. G. Leal, J. A. Jardini, L. C. Magrini, S. U. Ahn, Distribution
Transformer Losses Evaluation, A New Analytical Methodology and
Artificial Neural Network Approach, IEEE Transactions on Power
System, Vol. 24, No. 2, MAY 2009.
Copyright BAJECE
ISSN: 2147-284X
[21]
[22]
[23]
[24]
DOI: 10.17694/bajece.39364
96
M. Tripathy, R. P. Maheshwari and H.K. Verma, Probabilistic neuralnetwork-based protection of power transformer,IET Electr. Power
Appl. 1, (5), pp. 793798, 2007.
A. Stadler, M. Albach, S. Chromy, The Optimization of High
Frequency Operated Transformers for Resonant Converters, Dresden
EPE , 2005
O. kizli, Distribution, Losses and Warm up on Electrical Machines Elektrik Makinalarnda Dalma, Kayplar ve Isnma, Library of
Istanbul Technical University, 1962, No 482, pp. 15-18
M. Tsili, A. Kladas, P. Georgilakis, A. Souflaris, D. Paparigas,
Numerical techniques for design and modeling of distribution
transformers, Journal of Materials Processing Technology, 161. 320
326, 2005.
J. Kondoh, I. Ishii, Fault Current Limiting Transformer With Variable
Reactance, IEEE Transactions on applied superconductivity, Vol. 14,
No. 2, June 2004.
Jan Reynders, The prediction of fault currents in a large multi-winding
reactor transformer, IEEE Bologna PowerTech Conference, June 2326, Bologna, Italy, 2003.
L. Fausett, Fundamentals of Neural Networks: Architectures, Algorithms
and Applications, Prentice Hall. 1994.
H. Yurtoglu, Yapay Sinir Alar Metodolojisi ile ngr Modellemesi:
Baz Makroekonomik Deikenler iin Trkiye rnei, DPT Dissertion,
DPT:2683, pp 1-7. Feb. 2005
H. S. Nogay, Prediction of Internal Temperature in Stator Winding of
Three-Phase Induction Motors with ANN, European Transactions on
Electrical Power. 20:19. DOI: 10.1002. 2010.
T. Kohonen, State Of The Art In Neural Computing, IEEE First
International Conference on Neural Networks, 1. 79-90, 1987.
Z. Aydogmus, A neural network-based estimation of electric fields
along high voltage insulators, Expert Systems with Applications, 36,
87058710. 2009.
T.M. Hagan, H.B. Demuth, M. Beale, Neural Network Design, PWS
Publishing Company, Boston 2-44, 1996.
B. Ciylan, Determination of Output Parameters of a Thermoelectric
Module using Artificial Neural Networks, Electronics and Electrical
Engineering. Kaunas: Technologija, no:10, 116, 2011
G.E.P Box., G. Jenkins, Time Series Analysis, Forecasting and Control,
Golden-Day, San Francisco, CA, 1970.
M. Islam, A. Sattar, F. Amin, X. Yao, and K. Murase, A New
Constructive Algorithm for Architectural and Functional Adaptation of
Artificial Neural Networks, IEEE Transactions on Systems , Man. And
Cybernetics Part B, Vol. 39, No 6, Dec. 2009.
B. Jena, D. Swain, and A. Tyagi, Application of Artificial Neural
Networks for Sea-Surface Wind-Speed Retrieval From IRS-P4,
(MSMR), Brightness Temperature IEEE Geoscience and Remote
Sensing Letters, Vol. 7, No. 3, 567. July 2010.
Khaled Shaban, Ayman El-Hag, Andrei Matveev, A Cascade of
Artificial Neural Networks to Predict Transformers Oil Parameters,
IEEE Transactions on Dielectrics and Electrical Insulation ,Vol. 16, No.
2; April 2009.
A. Kazemi, M. T. Hassanzadeh, A. Gholami, Artificial Neural Network
for Insulator Leakage Currents Prediction from Environmental Data,
2nd IEEE International Conference on Power and Energy (PECon 08),
Johor Baharu, Malaysia, December 1-3, 2008
T.C. Akinci. Short Term Wind Speed Forecasting with ANN in
Batman, Turkey , Electronics and Electrical Engineering, Vol.107,
No.1, 2011.
BIOGRAPHY
SELCUK NOGAY was born in Isparta, Turkey. He

received the B.S. and M.S degrees from the
University of Marmara, Istanbul, Turkey. Since
2015, she is a Lecturer with the Erciyes University.
Her research interests are in Electrical Machines,
Renewable Energy and Artificial Intelligence.
September 2016
Vol.4 No.2
ISSN: 2147- 284X

Vol: 4
No:2
Year: September 2016
CONTENTS
R.Y. Kazakova, S.I. Nedelcheva, and R.K. Popov; Estimation of Wind Turbine
Generator Model Parameters using Artificial Intelligence Methods,...... . 51-57
D. Turkpence; Simulation of Entanglement Dynamics of Flying Qubits Through Photonic
Fields, .. 58-61
K.Dawood; Hybrid wind-solar reliable solution for Turkey to meet electric demand,
...........62-66
S. Singh and S. Bhullar; Hardware implementation of auto switching and light intensity
control of LED lamps, .................................................................................................. 67-71
D.K. Nedev, and K.D. Nedev; Aeronautical telecommunication simulator,
...72-74
S. Hansun; A New Approach of Browns Double Exponential Smoothing Method in Time
Series Analysis..... 75-78
C. Bakir; Automatic Speaker Gender Identification for the German Language,
...... 79-83
G. Kavuran, B. B. Alagoz, A. Ates, C. Yeroglu; Implementation of Model Reference
Adaptive Controller with Fractional Order Adjustment Rules for Coaxial Rotor Control
Test System, ..84-88
H.S. Nogay; Determination of Leakage Reactance in Monophase Transformers Using by
Cascaded Neural Network, ...........................................................................................89-96
BALKAN JOURNAL OF
ELECTRICAL & COMPUTER ENGINEERING
(An International Peer Reviewed, Indexed and Open Access Journal)
Contact
Department of Electrical Engineering,
Ayazaga Campus, Maslak, Istanbul-Turkey
Web: https://www.bajece.com
http://dergipark.ulakbim.gov.tr/bajece/
e-mail: editor@bajece.com
bajece.editor@gmail.co

Bajece Vol-4 No-2 Full

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bajece Vol-4 No-2 Full

Uploaded by

Copyright:

Available Formats

BAJECE

ISSN: 2147- 284X

R.Y. Kazakova, S.I. Nedelcheva, and R.K. Popov;

Aim & Scope

Istanbul Technical University

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

Estimation of Wind Turbine Generator Model

Abstract Modelling (in a broad sense) is an essential tool for

main focus in system identification is on the parameter

most powerful AI techniques is genetic algorithm (GA),

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

improves over generation sand the best solution is finally

The power of GA lies in its simultaneous searching a

PSO, the potential solutions, called particles, fly through the

v[ ] is the particle velocity, persent[ ] is the current particle

while maximum iterations or minimum error criteria is not

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

III. MODEL DESCRIPTION

For a description of processes in the synchronous generator

voltage, resulting from the relative rotation of the actual stator

- Mechanical output power of the turbine (W);

Turbine Power Characteristics (Pitch angle beta = 4 deg)

is the modulus of firmness of the asynchronous

Fig. 1. Power characteristics

Equation (6) can be normalized. In the per unit (pu) system

B. The model structure

A generic equation is used to model cp (,). This equation,

Fig. 2. The structural scheme of the mathematical model of the wind

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

model to be identified: Rsum , Rf , L12, L1, J . Monte Carlo

Crossover_p Crossover probability parameter;

Varried parameters in PSO algorithm are chosen to be:

Fig. 3. The structural scheme of the synchronous generator mathematical

For a description of the wind turbine a library model of

Correction_factor- is the correction (learning)

In the next step central composite design is performed. Only

Generator excitation voltage = 50 V;

The simulation results are illustrated in Fig.4

The position of the factors in the space of the parameters in GA is

Fig. 4. Simulation of change of currents along the axes d and q

Fig.6. Positions of the input variables in the parameter space in GA tests.

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

Central composite design for the PSO test parameters variation

Each algorithm was evaluated 90 times with varied parameters.

Scatter plot AvSAE vs. Crossover parameter and Mutation rate

A, B, AvSAE and err are the same as in function ga_proc;

The real coded GA search algorithm uses BLX- crossover

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

Sorted AvSAE in GA and PSO

Scatter plot AvSAE vs. Correction factor and Inertia

Provided null hypothesis F-test for each of tested

The next Fig.10 is showed sorted values of AvSAE for both

Test duration for all 90 estimations in each case (GA and

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

STEFKA NEDELCHEVA is Professor, PhD Eng.

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING,

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING,

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING,

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING,

BALKAN JOURNAL OF ELECTRICAL & COMPUTER ENGINEERING

Hybrid wind-solar reliable solution for Turkey

Abstract Energy is the most important part of life, it is

NERGY is one of the crucial inputs for socio-economic

K. DAWOOD, is with Department of Electrical Engineering University of