Doğrusal parametrik ve doğrusal olmayan gerçek sistemlerin yapay arı kolonisi algoritması kullanılarak modellenmesi
Öz
Araştırmacılar, klasik optimizasyon tekniklerinin kullanılmasının yetersiz olduğu problemler için farklı
çözüm arayışlarına girmiş, çalışmalarını bu yönde devam ettirmişlerdir. Yapay zeka optimizasyon teknikleri
ise bu çalışmaların bir sonucu olarak ortaya çıkmıştır. Yapay zeka optimizasyon teknikleri, karmaşık
optimizasyon problemlerinin çözümünde sık sık kullanılmaya başlanmış ve olumlu sonuçlar verdiği
görülmüştür. Doğadan esinlenen algoritmaların yeni bir dalı olan sürü zekâsı yaklaşımı, böceklerin
içgüdüsel problem çözme becerilerini kullanan etkili metasezgisel yöntemler geliştirebilmek için böcek
davranışlarının modellenmesine odaklanmıştır. Yapay Arı Kolonisi (YAK) Algoritması da sürü zekasına
sahip olan arıların davranışlarını esas alıp geliştirilmiş ve karmaşık problemlerin çözümünde kullanılmaya
başlanmıştır.
Bu çalışmada da, doğrusal olan parametrik ve doğrusal olmayan gerçek sistemlerin sürü zekası yaklaşımına
örnek olan yapay zeka optimizasyon tekniklerinden Yapay Arı Kolonisi (YAK) Algoritması ile modellenmesi
gerçekleştirilmiş olup, Genetik algoritma (GA) ve Klonal Seçme Algoritması (KSA) ile başarımı
karşılaştırılmıştır. Benzetim çalışmalarında literatürde kıyaslama problemlerinde sıklıkla kullanılan bir adet
doğrusal parametrik ve bir adet doğrusal olmayan gerçek sistem bu algoritmalar yardımıyla
modellendirilmiştir. Benzetim sonuçlarına göre doğrusal parametrik sistemlerin modellenmesinde YAK
algoritması, GA’ya çok yakın bir sürede parametre tahmini yapmış, farklı koloni sayılarında ise hem
GA’dan hemde KSA’dan daha düşük modelleme hatası ile sonuç alındığı görülmüştür. Doğrusal olmayan
gerçek sistem modellemelerde ise aynı durum söz konusu olup, koloni sayılarındaki değişime bağlı olarak
YAK algoritması, GA ve KSA’na göre daha düşük hata ve daha erken sürede sistemi modelleyebildiği
görülmüştür.
Anahtar Kelimeler
Modelling of linear parametric and non-linear real systems by using artificial bee colony algorithm
Abstract
A system can be define as in accordance with a
specific purpose in response to inputs producing
outputs that have a reciprocal interaction between
one element to another and the relation between
external world and within its elements. The main
purpose of using the term of the system is to
investigate the structure of the systems in order to
attain desired outcomes, to determine the basic
principles concerning with the system and to
regulate the system based on purposes if possible.
However, it is not always possible to investigate the
actual system and to determine its principles.
Therefore, a tool is needed to represent all the
features of the actual system. This tool will provide
the best way to understand the system and its
process . This representation tool, which will be
used for this purpose, is called model. In other
words, the model refers to simplified structures that
can usually be either mathematical or computable.
The purpose of modelling is to determine the
relation between input and output of an unknown
system. In other words, modelling aims to find
parameters of the transfer function that is
characterized by the system. Modelling should be
based on well defining for determination of complex
parameters. By all means, if a model is established
with chosen correct relationships, the solution would
lead to more accurate and better results.
A trend towards the use of the natural simulations is
increasing in order for modelling and solving
complex optimization problem as day goes on
because classical optimization algorithms are not
sufficient to solve the problems that oversized, linear
and non-linear mathematical or real system.
Modelling of real systems that suits a particular
solution method is often not easy. Nature-inspired
heuristic optimization algorithms, which are
independent from the problem and model, are
suggested to overcome deficiencies of conventional
optimization techniques.
New ways of searching have brought along in cases
where the use of classical optimization techniques is
insufficient. Artificial intelligence optimization
techniques have been proposed as a result of this
search.
Swarm intelligence approach, which is a new
branch of algorithms inspired by nature, is used
instinctively problem solving skills of the insects.
This approach has focused on the modeling of
insects' behavior to develop effective meta-heuristics
methods. An example of swarm intelligence
approach is ABC algorithm developed by Karaboga.
As a result of the interaction between insects, one of
the most important parts of collective intelligence is
to share information among insects individually.
Oscillation dance of honey bees can be given as an
example of types of interactive behaviors in which
honey bees share information regarding the source
and the quality of the discovered foods. By
performing the dance, honey bees give messages to
other bees in respect to quality of food supply, the
direction of the food, the distance and the amount of
the nectar. Through this successful mechanism, the
colony can be directed to the region where good
quality of food resources is available.
This study aims to determine the parameters
estimation by using Artificial Bee Colony (ABC)
Algorithm, Genetic Algorithm (GA) and Clonal
Selection Algorithm (CSA). The obtained results
have been compared with one algorithm to another.
Furthermore, Artificial Bee Colony (ABC)
Algorithm, which have been introduced into the
literature newly, has been given as a good example
of swarm intelligence approach.
In the literature, for comparison problems in the
modelling studies, one linear parametric and one
non-linear real system are often modelled by using
these algorithms. The results of modelling have
demonstrated that in the modelling of linear
parametric system, ABC algorithm estimated
parameter in a time very close to GA and indicated
less modelling error in the number of different
colonies than both GA and CSA. Same case also
occurs in the modeling of non-linear parametric
real system and depending on changes in the
number of colonies, ABC algorithm modelled the
system with less error in a shorter time compare to
GA and CSA.
Keywords
