Optimizasyon problemleri için yeni metasezgisel yaklaşımlar
Loading...
Date
2023-03-01
Authors
Kuyu, Yiğit Çağatay
Journal Title
Journal ISSN
Volume Title
Publisher
Bursa Uludağ Üniversitesi
Abstract
Optimizasyon işlemleri, mühendislik problemlerinde en sık karşılaşılan uygulamalardır. Bu tip problemlerin çözümünde klasik yöntemlerin yanında metasezgisel algoritmalardan da yoğun şekilde yararlanılmaktadır. Bu amaç doğrultusunda da birçok metasezgisel algoritma geliştirilmiş ve geliştirilmeye de devam edilmektedir. Bu tez çalışmasında; geometrik sekizli bölge arama algoritması (GSBA), modifiye adli tıp temelli soruşturma algoritması (modFBI) ve modifiye hiyerarşik yığın tabanlı optimizasyon algoritması (HBO-CO) olarak adlandırılan üç adet metasezgisel algoritma önerilmiştir. GSBA, böl ve yönet prensibinden ilham alırken, Cauchy tabanlı mutasyon ve karşıtlık temelli öğrenme diğer iki algoritmaya adapte edilmiştir. Önerilen metasezgisel algoritmalar, iki problem setine uygulanmış ve karşılaştırmalı biçimde test edilerek ayrıntılı performans değerlendirmeleri (Wilcoxon, Friedman testleri vb.) gerçekleştirilmiştir: birincisi sayısal (tek modlu, çok modlu ve yüksek boyutlu) fonksiyonlar ve ikincisi de 22 tane gerçek dünya problemleridir. Karşılaştırmalarda 5’i klasik ve 7’si güncel olmak üzere 12 tane algoritma kullanılmıştır. Gerçek dünya problemlerinde özellikle Elektrik-Elektronik mühendisliği uygulama alanına ait problemler (filtre tasarımları, harmonik eliminasyonları, dinamik ve statik ekonomik yük dağıtımları, parametre kestirimi, anten dizisi tasarımı vb.) ilgili algoritmalara uyarlanmış ve uygulanmıştır. Gerçekleştirilen testlerde önerilen GSBA algoritması, her iki problem setinde istatistikî olarak başarılı sonuçlara ulaşmıştır. Önerilen diğer iki algoritmanın; HBO-CO ve modFBI, nümerik sonuçlara göre kıyaslanan algoritmalardan daha iyi olduğu gözlemlenmiştir. Diğer yandan, gerçek dünya problemlerinde, bu üç algoritma arasında baskın bir algoritma olmayıp, bir problemde en iyi performansı gösteren algoritma diğer bir problemde en iyi olmayabilmektedir.
Optimization techniques are among the prevalent methods utilized in resolving engineering challenges. In addition to traditional techniques, the application of metaheuristic algorithms has become widespread in addressing complex problem scenarios. To this end, numerous metaheuristic algorithms have been devised and ongoing efforts are being made to enhance their efficacy. In this thesis, three metaheuristic algorithms called geometric octal zone distance estimation algorithm (GSBA), modified forensic-based investigation algorithm (modFBI) and modified hierarchical heap-based optimization algorithm (HBO-CO) are proposed. The GSBA is inspired by the divide-and-conquer principle, while Cauchy-based mutation and opposition-based learning are adapted to the other two algorithms. The proposed algorithms are implemented and tested on two sets of problems, with a subsequent comparison of their performance. A comprehensive assessment of their performance is conducted (Wilcoxon, Friedman tests, etc.) through two distinct evaluation approaches. The first is numerical (unimodal, multimodal and high-dimensional) functions and the second is 22 real-world problems by comparing with 12 algorithms in total, 5 classical and 7 up-to-date. In real world problems, this study also incorporates problems relevant to the domain of Electrical and Electronics Engineering applications (filter designs, harmonic eliminations, dynamic and static economic load distributions, parameter estimation, antenna array design, etc.). From the experimental analysis, GSBA has achieved the most successful results for both numerical problem sets according to statictical analyzes. The other two proposed algorithms, which are HBO-CO and modFBI, were observed to be better than or at least comparable to the compared algorithms according to numerical results. On the other hand, in real-world problems, there is no absolute leader among the developed three algorithms. It can be observed that the algorithm that performs best in one problem cannot be the best in another problem.
Optimization techniques are among the prevalent methods utilized in resolving engineering challenges. In addition to traditional techniques, the application of metaheuristic algorithms has become widespread in addressing complex problem scenarios. To this end, numerous metaheuristic algorithms have been devised and ongoing efforts are being made to enhance their efficacy. In this thesis, three metaheuristic algorithms called geometric octal zone distance estimation algorithm (GSBA), modified forensic-based investigation algorithm (modFBI) and modified hierarchical heap-based optimization algorithm (HBO-CO) are proposed. The GSBA is inspired by the divide-and-conquer principle, while Cauchy-based mutation and opposition-based learning are adapted to the other two algorithms. The proposed algorithms are implemented and tested on two sets of problems, with a subsequent comparison of their performance. A comprehensive assessment of their performance is conducted (Wilcoxon, Friedman tests, etc.) through two distinct evaluation approaches. The first is numerical (unimodal, multimodal and high-dimensional) functions and the second is 22 real-world problems by comparing with 12 algorithms in total, 5 classical and 7 up-to-date. In real world problems, this study also incorporates problems relevant to the domain of Electrical and Electronics Engineering applications (filter designs, harmonic eliminations, dynamic and static economic load distributions, parameter estimation, antenna array design, etc.). From the experimental analysis, GSBA has achieved the most successful results for both numerical problem sets according to statictical analyzes. The other two proposed algorithms, which are HBO-CO and modFBI, were observed to be better than or at least comparable to the compared algorithms according to numerical results. On the other hand, in real-world problems, there is no absolute leader among the developed three algorithms. It can be observed that the algorithm that performs best in one problem cannot be the best in another problem.
Description
Keywords
Optimizasyon, Metasezgisel algoritma, Gerçek dünya problemi, Parameter estimation, Metaheuristic algorithm, Real-world problem
Citation
Kuyu, Y. Ç. (2023). Optimizasyon problemleri için yeni metasezgisel yaklaşımlar. Yayınlanmamış doktora tezi. Bursa Uludağ Üniversitesi Fen Bilimleri Enstitüsü.