You are on page 1of 10

1

Using Imperialist Competitive Algorithm to reduce


the error between RBF and Training Data with
MATLAB
Jose Eduardo Urrea Cabus, Mohammad Baghdadi
Karadeniz Technical University
Faculty of Engineering
Department of Electrical and Electronics Engineering

Abstract—Solving systems of nonlinear equations is a relatively chemistry, business, industry, engineering and computer sci-
complicated problem in which arise a diverse range of sciences. ence. All the countries are divided into two types: imperialist
There are a number of different approaches that have been states and colonies. Imperialistic competition is the main part
proposed. In this paper, we employ the imperialist competitive
algorithm (ICA) for solving systems of nonlinear equations. Some of proposed algorithm and hopefully cause the colonies y/o
well-known problems are presented to demonstrate the efficiency converge to the global minimum of the cost function [1].
of this new robust optimization method in comparison to other
known methods [2]. The Imperialist Competitive Algorithm In computer science, imperialist competitive algorithm is a
(ICA), derived from the field of human social evolution, is a computational method that is used to solve optimization prob-
component of swarm intelligence theory. It was first introduced
lems of different types. Like most of the methods in the area
in 2007 to deal with continuous optimization problems, but
recently has been extensively applied to solve discrete opti- of evolutionary computation, ICA does not need the gradient
mization problems [3]. The proposed algorithm start with an of the function in its optimization process. From a specific
initial population. Population individuals called country are in point of view, ICA can be thought of as the social counterpart
two types: colonies and imperialists that all together form some of genetic algorithms (GAs). ICA is the mathematical model
empires. Imperialistic competition among these empires forms
and the computer simulation of human social evolution, while
the basis of the proposed evolutionary algorithm. During this
competition, weak empires collapse and powerful ones take GAs are based on the biological evolution of species [6].
possession of their colonies. Imperialistic competition hopefully
converges to a state in which there exist only one empire and The Imperialist Competitive Algorithm (ICA), a recently
its colonies are in the same position and have the same cost developed metaheuristic introduced by Atashpaz-Gargari and
as the imperialist. Applying the proposed algorithm to some Lucas, was inspired by socio-political behaviors. The initial
of benchmark cost functions, shows its ability in dealing with
different types of optimization problems [1].
work on the ICA was dedicated to the continuous optimization
problems, but it is currently applied to many complex dis-
Index Terms—ICA, GA, algorithm, imperialist, country, crete combinatorial optimization problems, such as Flowshop
colonies, MSE, RMSE, MNN, RWS.
Scheduling Problem (FSP), Assembly Line Balancing Problem
(ALBP), Traveling Salesman Problem (TSP), and Facility
I. I NTRODUCTION Layout Problem (FLP) [3].

F rom a general point of view, optimization is the process


of making something better. Having a function f (x) in
optimization, we want to find an argument x whose relevant
II. A HISTORICAL REVIEW OF IMPERIALISM
Imperialism is the policy of extending the power and rule
cost is optimum (usually minimum). Different methods have of a government beyond its own boundaries. A country may
been proposed for solving an optimization problem. Some attempt to dominate others by direct rule or by less obvious
of these methods are the computer simulation of the natural means such as a control of markets for good or raw materials.
processes. For example genetic algorithms GA are a particular The latter is often called neocolonialism. In its initial forms,
class of evolutionary algorithms that evolve a population imperialism was just a political control over other countries in
of candidate solutions to a given problem, using operators order to only use their resources. Also in some cases the reason
inspired by natural genetic variation and natural selection. to control another country was just preventing the opponent
imperialist from taking possession of it. No matter what the
The available optimization algorithms are extensively used reason was, the imperialist states were competing strongly
to solve different optimization problems such as industrial for increasing the number of their colonies and spreading
planning, resource allocation, scheduling, decision making, their empires over the world. This competition resulted in
pattern recognition and machine learning. Furthermore, op- a development of the powerful empires and the collapse
timization techniques are widely used in many fields such as of weaker ones. In the proposed algorithm, the imperialists
Jose Eduardo Urrea Cabus, (email: joseeduardourrea@gamil.com). do the same for their colonies. Here, all the imperialists
Mohammad Baghdadi, (email: mohammad.baghi93@gmail.com). compete for taking possession of colonies of each other. Also
2

assimilation policy is modeled by moving the colonies toward characteristics (optimization search space). Revolution brings
the imperialists [1]. about sudden random changes in the position of some of
the countries in the search space. During assimilation and
Developed countries of today attempt to control less- revolution a colony might reach a better position and has the
developed countries through direct means, such as direct chance to take the control of the entire empire and replace the
legislation, or through indirect means, such as capturing their current imperialist state of the empire.
markets through exports and controlling their imports. This
type of imperialism is called neocolonialism. Within optimiza- Imperialistic Competition is another part of this algorithm.
tion theory, the colonialism phenomenon lifts a dominating All the empires try to win this game and take possession of
colony out of a valley (current position) and moves it toward colonies of other empires. In each step of the algorithm, based
an imperialist peak (new minimum area). The new position on their power, all the empires have a chance to take control of
of colony might even be better than the imperialist at any one or more of the colonies of the weakest empire. Algorithm
given time. Fig. 1 explicitly shows a colony that migrated continues with the mentioned steps (Assimilation, Revolution,
from a minimum area to a new minimum area along the Competition) until a stop condition is satisfied [6].
economy axis. Movement in the economy axis means that
the colony improves its economy situation by taking influence IV. THE PROPOSED ALGORITHM
of imperialist’s economy. This process is called assimilation Figure 2 shows the flowchart of the proposed algorithm.
process. Like other evolutionary ones, the proposed algorithm starts
with an initial population (countries in the world). Some
of the best countries in the population are selected to be
the imperialists and the rest form the colonies of these
imperialists. All the colonies of initial population are divided
among the mentioned imperialists based on their power. The
power of an empire which is the counterpart of the fitness
value in GA, is inversely proportional to its cost.

After dividing all colonies among imperialists, these


colonies start moving toward their relevant imperialist country.
The total power of an empire depends on both the power of
the imperialist country and the power of its colonies. Then the
imperialistic competition begins among all the empires. Any
Figure 1. Assimilating a colony toward an imperialist empire that is not able to succeed in this competition and can-
not increase its power (or at least prevent decreasing its power)
Moving colony toward its imperialist could result in better will be eliminated from the competition. The imperialistic
solution as represented in Fig. 1. According to Fig. 1, during competition will gradually result in an increase in the power
assimilation process, the colony from its current position (point of powerful empires and a decrease in the power of weaker
A) moves toward its imperialist (point C) and placed in a ones. Weak empires will lose their power and ultimately they
new position (point B). Fig. 1 also shows that the assimilated will collapse. The movement of colonies toward their relevant
colony explores new search space, which is global optimum. imperialists along with competition among empires and also
[3]. the collapse mechanism will hopefully cause all the countries
to converge to a state in which there exist just one empire in the
III. THEORETICAL BACKGROUND world and all the other countries are colonies of that empire.
This algorithm starts by generating a set of candidate In this ideal new world colonies, have the same position and
random solutions in the search space of the optimization power as the imperialist [1].
problem. The generated random points are called the initial
Countries. Countries in this algorithm are the counterpart A. Generating Initial Empires (Initialization)
of Chromosomes in GAs and Particles in Particle Swarm
Optimization (PSO) and it is an array of values of a candidate The goal of optimization is to find an optimal solution in
solution of optimization problem. The cost function of the terms of the variables of the problem. We form an array of
optimization problem determines the power of each country. variable values to be optimized. In GA terminology, this array
Based on their power, some of the best initial countries is called chromosome, but here the term country is used for
(the countries with the least cost function value), become this array. In an Nvar −dimensional optimization problem, a
Imperialists and start taking control of other countries (called country is a 1xNvar array. This array is defined by
colonies) and form the initial Empires.
country = [p1 , p2 , p3 , · · · , pNvar ] (1)
Two main operators of this algorithm are Assimilation and
Revolution. Assimilation makes the colonies of each empire The variable values in the country are represented as floating
get closer to the imperialist state in the space of socio-political point numbers. The cost of a country is found by evaluating
3

Cn = cn − max(ci ) (3)

Where cn is the cost of n0 th imperialist and Cn is its nor-


malized cost. Having the normalized cost of all imperialists,
the normalized power of each imperialist is defined by

Cn



Pn = NP
imp (4)
Ci

i=1

From another point of view, the normalized power of an


imperialist is the portion of colonies that should be possessed
by that imperialist. Then the initial number of colonies of an
empire will be

N.C.n = round{pn .Ncol } (5)

Where N.C.n is the initial number of colonies of n0 th


empire and Ncol is the number of all colonies. To divide the
colonies, for each imperialist we randomly choose N.C.n of
the colonies and given them to it. These colonies along with
the imperialist will form n0 th empire. The figure 3 shows the
initial population of each empire. As shown in this figure
bigger empires have greater number of colonies while weaker
ones have less. In this figure imperialist 1 has formed the most
powerful empire and has the greatest number of colonies [1].

Figure 2. Flowchart of the proposed algorithm

Figure 3. A representation of forming empires. The more colonies an


the cost function f at the variables (p1 , p2 , p3 , · · · , pNvar ). imperialist possess, the bigger is its relevant start mark
Then
B. Assimilation process(moving colonies toward imperialist)
cost = f (country) = f (p1 , p2 , p3 , · · · , pNvar ) (2)
As part of the colonial movement, imperialists build infras-
To start the optimization we generate the initial population tructures such as transportation systems within their colonies;
of size Npop . We select Nimp of the most powerful countries thus attempting to assimilate colonies [3]. In in other words,
to form the empires. The remaining Ncol of the population will imperialists countries started to improve their colonies. The
be the colonies each of which belongs to an empire. Then we movement is show in the figure 4 in which the colony moves
have to types of countries: imperialist and colony. toward the imperialist by x units. The new position of colony
is shown in a darker color. The direction of the movement
To form the initial empires, we divide the colonies among is the vector from colony to imperialist. In this figure x is
imperialists based on their power. That is the initial number a random variable with uniform (or any proper) distribution.
of colonies of an empire should be directly proportionate to Then for x we have
its power. To divide the colonies among imperialists propor-
tionally, we define the normalized cost of an imperialist by x ∼ U (0, β × d) (6)
4

Where β is a number greater than 2 and d is the distance increases exploration and prevents the early convergence of
between colony and imperialist. A β > 1 causes the colonies countries to local optima.
to get closer to the imperialist state from both sides.
Figure 6 shows the key role of revolution. As it can be
seen from Figure 6, the colony located at point A can be
transferred to some point between point A and B. It is obvious
that assimilation process, in this case, may cause colony to get
trapped to local optimum. By performing revolution operation,
colony suddenly jumps from point A to point C which results
in escaping from local optimum [3].

Figure 4. Moving colonies toward their relevant imperialist

To search different points around the imperialist we add a


random amount of deviation to the direction of movement.
Figure 5 shows the new direction. In this figure θ is a random
number with uniform (or any proper) distribution. Then
θ ∼ U (−γ, γ) (7)

Where γ is a parameter that adjusts the deviation from the Figure 6. Colony escapes from local optimum by performing revolution
original direction. Nevertheless the values of β and γ are operation.
arbitrary, in most of our implementation a value of about 2
for β about π/4 for γ have resulted in good convergence of
countries to the global minimum. The number of colonies of the empire which is supposed to
be replaced with the number of new generated countries is:

N.R.C = round{µ × Ncol } (8)


Where µ is the Revolution Rate, Ncol is
N o.(the colonies of empiren ) and N.R.C is the number
of revolutionary colonies. This will improve the global
convergence on the ICA and prevent it sticking to a local
minimum [2].

D. Exchanging positions of the imperialist and a colony

While moving toward the imperialist, a colony may reach


to a position with lower cost than that of imperialist. In such a
case, the imperialist moves to the position of that colony and
vise versa. Then the algorithm will continue by the imperialist
Figure 5. Moving colonies toward their relevant imperialist in a randomly
deviated direction in a new position and then colonies start moving toward this
position, Figure 7.
C. Revolution
Figure 8 depicts the position exchange between a colony and
According to imperial history, colonies of an empire are the imperialist. In this figure the best colony of the empire is
absorbed by imperialist in terms of social, cultural, economic, shown in a darker color. This colony has a lower cost than
and political characteristics; however, there might be some that of imperialist. Figure 9 shows the whole empire after
colonies that resist to be absorbed by imperialists. In fact, exchanging the position of the imperialist and that colony [1].
those colonies perform some sort of reformations in their
characteristics. In ICA, this operation is called revolution. E. Total power of an Empire
Revolution brings sudden random changes in the position of
some the colonies in the search space. Revolution operation Total power of an empire is mainly affected by the power
in ICA is comparable to mutation operation in GA, which of imperialist country. But the power of the colonies of an
5

Figure 9. The entire empire after position exchange

Figure 7. Exchanging position of imperialist and colony.

Figure 8. Exchanging the positions of a colony and the imperialist


Figure 10. Imperialistic competition. The more powerful an empire is the
more likely and it will possess the weakest colony of the weakest empire
empire has an effect, albeit negligible, on the total power of
that empire. It is represented as the total cost by
Where T.C.n and N.T.C.n are respectively total cost and
T.C.n = cost(imperialistn ) + ξmean{Ψ} (9) normalized total cost of n0 th empire. Having the normalized
total cost, the possession probability of each empire is given
Where Ψ is Cost(colonies of empiren ), T.C.n is the total by.
cost of the n0 th empire and ξ is a positive number which is
N.T.C.n

considered to be less than 1. A little value for ξ causes the total N
power of the empire to be determined by just the imperialist Pn = P imp (11)
and increasing it will increase the role of the colonies in N.T.C.i
i=1
determining the total power of an empire. We have used the
value 0.1 for ξ in most of our implementation. To divide the mentioned colonies among empires on the
possession probability of them, we form the vector P as
P=[Pp1 , Pp2 , Pp3 , · · · , PpNimp ].
F. Imperialistic competition
All the empires try to take possession of colonies of Then we create a vector with the same size as P
other empires and control them. This imperialistic competition whose elements are uniformly distributed random numbers,
gradually brings about a decrease in the power of weaker R=[r1 , r2 , r3 , · · · , rNimp ], r1 , r2 , r3 , · · · , rNimp ∼ U (0, 1).
empires and an increase in the power of more powerful ones.
This competition modeled by picking some of the weakest Then we form vector D by simply subtracting R from P,
colonies of the weakest empires and making a competition D=P-R. Referring to vector D we will hand the mentioned
among all empires to possess these (this) colonies. Figure 10 colonies to an empires whose relevant index in D is maximum.
shows a big picture of the modeled imperialistic competition.
Based on their total power, in this competition, each of empires G. Eliminating the powerless empires
will have a likelihood of taking possession of the mentioned
Powerless empires will collapse in the imperialistic compe-
colonies. In other words these colonies will not be possessed
tition and their colonies will be divided among other empires.
by the most powerful empires, but these empires will be more
In modeling collapse mechanism different criteria can be
likely to possess them.
defined for considering an empire powerless. In most of our
implementation, we assume an empire collapsed and eliminate
To start the competition, first, we find the possession proba- it when loses all of its colonies [1], figure 11.
bility of each empire based on its total power. The normalized
H. Convergence
total cost is simply obtained by
After a while all the empires except the most powerful one
N.T.C.n = T.C.n − max{T.C.i } (10) will collapse and all the colonies will be under the control of
6

2. Direction assimilation parameter(β)


Direction parameter β ensures how far colony gets close
to its imperialist during assimilation strategy. It plays an
important role to make a trade-off between exploitation and
exploration. A small value of β promotes exploitation, because
colony does move a little from its current position while a large
value of β results in exploration due to the large movement
Figure 11. Collapsing the weakest empire from its current position.

this unique empire. In this ideal new world all the colonies 3. Number of countries (N ), number of imperialist countries
will have the same positions and same costs and they will be (Nimp ), number of colonies (Ncol )
controlled by an imperialist with the same position and cost as Number of countries in the ICA is the counterpart of popu-
themselves. In this ideal world, there is no difference not only lation in the GA. It compromises the solution accuracy and
among colonies but also between colonies and imperialist [1], search time. Small number of countries may result in less
figure [12]. computational effort but poor solutions, while large number
of countries may result in better solution quality but high
computational efforts. In fact, the number of countries com-
promises the solution accuracy and search time and must be
determined based on the size and complexity of optimization
problems. The number of imperialist countries is usually
10–13% of the number of overall countries, and the remainder
would be number of colonies. For example, if we consider 80
countries for a particular problem, then it is better to start by
setting 8 (10%) countries as imperialists and 72 countries, the
Figure 12. Convergence representation of ICA
remainder, as colonies.

V. E FFECT OF ICA PARAMETERS TUNING ON SOLUTION 4. Coefficient associated with average power of empire’s
QUALITY colonies (ξ)
ξ is a positive constant for considering the average power
The performance of a metaheuristic algorithm is highly
of colonies in each empire. It is used to adjust the effect of
dependent on its parameter tuning. A better parameter tuning
cumulative power of colonies on determining the power of
allows larger flexibility and robustness, but requires careful ini-
each empire. A large value of ξ emphasizes more contribution
tialization. It is not obvious to define a priori which parameter
of power of colonies on the power of empire, while a small
setting should be used. It is true to say that the optimal values
value pinpoints that power of empire is almost entirely depen-
of the parameters of any metaheuristic mostly depend on the
dent on the power of its imperialist. Results reviewed in the
size and type of problem, and also on the search time that user
literature indicate that a small value of 0.1 or 0.15, could be
wants to spend for solving the problem. In the following, a
an appropriate choice.
brief description of adjustment of ICA parameters is summed
up based on the literature.
5. Revolution Rate (pr )
1. Deviation assimilation parameter (θ) Revolution operator is the counterpart of mutation in GA.
A large value of revolution rate reinforces the exploration,
Deviation parameter is one of the most critical parameters
while a small value of it may encourage exploitation. The
of ICA. It defines the compromise between search explo-
value of revolution rate is highly dependent on the magnitude
ration (diversification using more randomness) and search
of solution spaces; however the value of 0.2 could be an
exploitation (intensification using more greediness). In fact,
appropriate choice, in general [3].
small values for θ encourages exploitation, while large values
reinforces exploration. An appropriate way to adjust θ is
setting it to a large value at the beginning of search and 6. RBF Adaptive Control
gradually reducing to a small value at the end of search. This The RBF network adaptation can effectively improve the
can be done using a fuzzy adaptive engine or linear decreasing control performance against large uncertainty of the system.
functions. Notably, setting θ to a small value needs more The adaptation law is derived using the Lyapunov method,
search time, while a large value of θ investigates a good range hence the stability of the entire system and the convergence
of solution spaces (reduction in search time) but may sacrifice of the weight adaptation are guaranteed.
the solution accuracy. According to the literature, setting θ to
π/4 compromises between solution accuracy and search time In 1990, artificial neural networks were first proposed for
for most of the case studies. the adaptive control of nonlinear dynamical systems. Since that
time, both multilayer neural networks (MNN) and radial basis
7

function (RBF) networks have been used in numerous applica-


tions for the identification and control. RBF neural networks
were addressed in 1988, which have recently drawn much
attention due to their good generalization ability and a simple
network structure that avoids unnecessary and lengthy cal-
culation as compared to the multilayer feed-forward network
(MFN). Past research of universal approximation theorems on
RBF has shown that any nonlinear function over a compact set Figure 13. Parameters
with arbitrary accuracy can be approximated by RBF neural
network. There have been significant research efforts on RBF
VI. EXPERIMENT AND RESULTS
neural control for nonlinear systems [4].
The goal of optimization is to find an optimal solution in
terms of the variables of the problem. Taking as a reference
RBF neural network has three layers: the input layer, the above described our goal function is M SE (Mean Square
hidden layer, and the output layer. Neurons at the hidden Error) defined in the following equation using a MATLAB
layer are activated by a radial basis function. The hidden function:
layer consists of an array of computing units called hidden M SE = mean(e2 ) (16)
nodes. Each hidden node contains a center c vector that is a
parameter vector of the same dimension as the input vector And also we can get RMSE (Radical Mean Square Error) as
x; the Euclidean distance between the center and the network our goal function, as shown in the following equation:
input vector x is defined by: √ p
RM SE = M SE = mean(e2 ) (17)
||x(t) − cj (t)|| (12) Where e represent our error value. Hence, our error value can
be defined as the following equation
The output of hidden layer can be produced through a
e = y − yb (18)
nonlinear activation function hj (t) as follows:
Where y is the input proposed function and yb is our plot
||x(t) − cj (t)||2 function, plotted by using the RBF method and optimized with

2b2j ICA using n kernels (Gaussians). For the calculation we used
hj (t) = e , j = 1, · · · , m (13)
n = 6 kernels, where n can be increase and decrease to get
the best result. The following MATLAB code was used for
where bj notes a positive scalar called a width and m notes
the calculations of y and yb
the number of hidden nodes. The output layer is a linear
weighted combination as follows: for i =1: n K e r n e l
yhat=yhat+w( i ) * exp ( − 0 . 5 * ( ( x−m( i ) ) / sigma ( i ) ) . ˆ 2 ) ;
yy=yy+w( i ) * exp ( − 0 . 5 * ( ( xx−m( i ) ) / sigma ( i ) ) . ˆ 2 ) ;
m
X end
yi (t) = wji hj (t), i = 1, · · · , n (14)
j=1 The most important part of our algorithm, is assimilation
part and choosing the country’s and give them the position and
where w are the output layer weights, n notes the number cost and after that going to start the main loop of our algorithm
of outputs, and y notes the network output [4]. we selected a method to choose countries. Furthermore, we
choose RWS (Roulette Wheel Selection) as our method, shown
in the following MATLAB code:
7. Roulette Wheel Selection (RWS)
f u n c t i o n i=R o u l e t t e W h e e l S e l e c t i o n (P)
A common selection approach assigns a probability of selec- r=rand ;
tion Pj to each individual j based on its fitness value. A series C=cumsum (P) ;
i=f i n d ( r<=C, 1 , ' f i r s t ' ) ;
of N random numbers is generated P and compared against the end
cumulative probability Ci = Pi of the population. The
appropriate individual i is selected and copied into the new In this section, we minimized the error between our input
population if Ci − 1 < U [0 1] ≤ Ci .Various methods exist functions (F1 = y1 and F2 = y2 ) and yb1 and yb2 , respectively
to assign probabilities to individuals: roulette wheel, linear and plotting by using ICA. The following functions are:
ranking and geometric ranking. Roulette wheel, developed by F1 (x) = x2 − 4 and F2 (x) = x − 3. Our codes include 12
Holland [5], is the first selection method. The probability Pi parts, enumerate below:
for each individual is defined by: 1) AssimilateColonies
2) CreateData
Fi 3) CreateInitialEmpires
Pi = P opSize
(15)
P 4) DoRevolution
Fj
j=1
5) Ica
6) InterEmpireCompetition
Where Fi is fitness of individual i. 7) IntraEmpireCompetition
8

8) MyCost We got the following answers from the parameters shown


9) PlotSolution in Table II:
10) RouletteWheelSelection Iteration 1: Best Cost = 1020.2389
11) ShareSettings Iteration 2: Best Cost = 793.1481
12) UpdateTotalCost Iteration 3: Best Cost = 624.7888
Iteration 4: Best Cost = 467.2486
By changing our algorithm parameters we can control the Iteration 5: Best Cost = 461.2768
..
RBF parameters (σ, m and w) to reduce the error between y .
..
and yb. .
Iteration 380: Best Cost = 2.4231
For our first function F1 we have tested with the following Iteration 381: Best Cost = 2.4187
parameters in Table I and II to be analyzed. Iteration 382: Best Cost = 2.4131
Iteration 383: Best Cost = 2.4102
Iteration 384: Best Cost = 2.4089
Parameters Values
Iterations 10 Iteration 385: Best Cost = 2.4066
Kernel 2 Iteration 386: Best Cost = 2.403
Countries 50 Iteration 387: Best Cost = 2.4017
Empires 10 Iteration 388: Best Cost = 2.4001
Colonies 40 ..
β 2 .
θ 0 ..
ξ 0.1 .
α 1 Iteration 490: Best Cost = 2.3117
Table I Iteration 491: Best Cost = 2.3111
2
F1 (x) = x − 4
Iteration 492: Best Cost = 2.3103
Iteration 493: Best Cost = 2.3096
Iteration 10: Best Cost = 687.1832 Iteration 494: Best Cost = 2.3091
Iteration 495: Best Cost = 2.3079
Iteration 496: Best Cost = 2.3076
Iteration 497: Best Cost = 2.3072
Iteration 498: Best Cost = 2.3071
Iteration 499: Best Cost = 2.3067
Iteration 500: Best Cost = 2.3065

Figure 14. This graphic shows how the algorithm is following the points in
the function F1 (x) = x2 − 4 with 10 iterations and 2 kernels. Figure 15. Empires at the iteration 500, all the countries have converged to
the global minimum of the function being controlled by a unique empire.

Parameters Values As a conclusion of the comparison on graphic shown in


Iterations 500 figure 14 and 17 we can say that increasing the number of
Kernel 6
Countries 50
iterations and kernels, we can reduce the error between y1
Empires 10 and yb1 .
Colonies 40
β 2
θ 0
ξ 0.1
α 1
Table II
F1 (x) = x2 − 4
9

Figure 18. This graphic shows how the algorithm is following the points in
Figure 16. Best Cost vrs Iterations (500). the function F2 (x) = x − 3 with 10 iterations and 2 kernels.

Parameters Values
Iterations 500
Kernel 6
Countries 50
Empires 10
Colonies 40
β 2
θ 0
ξ 0.1
α 1
Table IV
F2 (x) = x − 3

Iteration 3: Best Cost = 1.9475


Iteration 4: Best Cost = 0.85082
Iteration 5: Best Cost = 0.85082
..
.
..
.
Figure 17. This graphic shows how the algorithm is following the points in Iteration 380: Best Cost = 0.0073477
the function F1 (x) = x2 − 4.
Iteration 381: Best Cost = 0.0073378
Iteration 382: Best Cost = 0.0073307
For our second function F2 we have tested with the follow- Iteration 383: Best Cost = 0.0073248
ing parameters in Table III and IV to be analyzed. Iteration 384: Best Cost = 0.0073181
Iteration 385: Best Cost = 0.0073096
Iteration 386: Best Cost = 0.0073016
Parameters Values Iteration 387: Best Cost = 0.0072899
Iterations 10 Iteration 388: Best Cost = 0.0072899
Kernel 2 ..
Countries 50 .
Empires 10 ..
.
Colonies 40
β 2 Iteration 490: Best Cost = 0.0065185
θ 0 Iteration 491: Best Cost = 0.0065139
ξ 0.1 Iteration 492: Best Cost = 0.0065099
α 1
Table III Iteration 493: Best Cost = 0.0065031
F2 (x) = x − 3 Iteration 494: Best Cost = 0.0064979
Iteration 495: Best Cost = 0.0064845
Iteration 496: Best Cost = 0.0064769
Iteration 10: Best Cost =0.084799 Iteration 497: Best Cost = 0.006466
Iteration 498: Best Cost = 0.0064591
We got the following answers from the parameters shown Iteration 499: Best Cost = 0.0064466
in Table IV: Iteration 500: Best Cost = 0.0063581
Iteration 1: Best Cost = 2.258
Iteration 2: Best Cost = 2.258
10

R EFERENCES
[1] Atashpaz-Gargari, E.; Lucas, C (2007). ”Imperialist
Competitive Algorithm: An algorithm for optimization in-
spired by imperialistic competition”. IEEE Congress on Evo-
lutionary Computation. 7. pp. 4661–4666.
[2] Mahdi Abdollahi, Ayaz Isazadeh, Davoud Abdollahi,
C(2013). ”Imperialist competitive algorithm for solving sys-
tems of nonlinear equations”. Computers and Mathematics
with Applications, Volume 65, Issue 12, August 2013, Pages
1894-1908.
[3] Seyedmohsen Hosseini, AbdullahAl Khaled. C(2014).
”A survey on the Imperialist Competitive Algorithm meta-
heuristic: Implementation in engineering domain and direc-
tions for future research”. Applied Soft Computing Volume
24, November 2014, Pages 1078-1094.
Figure 19. Empires at the iteration 500, all the countries have converged to [4] Radial Basis Function (RBF) Neural Network Control
the global minimum of the function being controlled by a unique empire.
For Mechanical Systems Design, Analysis And Matlab Simu-
lation, Jinkun Liu, ISBN 978-3-642-34815-0, Chapter1.
[5] Holland J., Adaptation in Natural and Artificial Systems,
University of Michigan Press, Ann Arbor, 1975.
[6] https://en.wikipedia.org/w/index.php?title=Imperialist
competitive algorithm&oldid=863176474

B IOGRAPHIES
Mr. José Eduardo Urrea Cabus
was born in Puerto Cortes,
Honduras, on January 02, 1992.
Received his B.Sc in Electrical
and Electronic Engineering
from the National Autonomous
University of Honduras in the
Figure 20. This graphic shows how the algorithm is following the points in
Sula Valley, Cortés, Honduras.
the function F2 (x) = x − 3. He has served as an Electrical
Engineer in the private sector
in Honduras for more than two
As a conclusion of the comparison on graphic shown in years. Now he is studying his
figure 18 and 20 we can say that increasing the number of MSc. degree in Electrical Engineering at Karadeniz Technical
iterations and kernels, we can reduce the error between y2 University.
and yb2 .
Mr. Mohammad Baghdadi was
VII. C ONCLUSION born in Gorgan, Iran, on January
This paper proposes an approach for solving systems of 27, 1987. Received his B.Sc. and
nonlinear equations. The system of nonlinear equations was M.Sc. in Power Electrical Engi-
transformed into a multi-objective optimization problem. The neering from Islamic Azad Uni-
goal was to obtain values as close to real values as possible for versity of Aliabad (A.I.A.U), Iran,
each of the involved objectives. Some well-known problems in 2011 and 2015, respectively.
were presented to demonstrate the efficiency of the Imperialist Now he is studying his PhD. in
Competitive Algorithm (ICA). By changing the values of ICA Electronic Engineering in Karad-
parameters (α, β, number of kernel, etc.) we can reduce eniz Technical University. He has
the error between yn and ybn , and changing our method to served as an Electrical Engineer in
chose our countries from RWS to SUS (Stochastic Universal the private sector in Iran for more
Sampling), we can control the parameters of RBF (σ, m and than 5 years. His thesis topic in
w) to reduce the error between RBF and training data. master was about develop optimal
sizing and locating of two SVC in distribution networks
ACKNOWLEDGMENT with dispersed photovoltaic by using imperialist competitive
The authors are very grateful to the Department of EEE, algorithm.
Karadeniz Technical University.