Publication: Enhancing fireworks algorithm for dynamic optimization problems
Abstract
Farklı alanlardaki birçok gerçek dünya probleminin dinamik karakteristikgöstermesi, son yirmi yılda araştırmacıların artan bir şekilde dinamik eniyilemeproblemleri üzerinde çalışmasında temel itici güç olmuştur. Bir dinamik eniyilemeprobleminde değişim zamanla olduğu için, bu tür bir problem için sunulan algoritmanınhedefi, zamanla değişen optimumu takip etmektir. Dinamik eniyileme problemleriniçözmek için literatürde evrimsel algoritmalar ve çeşitli sürü zekası teknikleri önerilmiştir.Havai Fişek Algoritması (FWA) son zamanlarda önerilmiş bir sürü zekası algoritmasıdır.Bu algoritma, havai fişeklerin patlamasını simüle ederek karmaşık statik problemleriçin global eniyilemeyi hedefler. Literatürde geleneksel Havai Fişek Algoritması üzerinebirçok iyileştirme sunulsa da Genişletilmiş Havai Fişek Algoritması (EFWA) bunlararasında en belirgin olanıdır. Bu tezde, gerçek uzayda yer alan dinamik eniyilemeproblemlerini çözmek üzere üç farklı EFWA tabanlı çözüm önerilmiştir. EFWA tabanlıçözümlerimizin performans değerlendirmesi Hareket Eden Tepeler (MPB) kıyaslamaproblemi ile doğrulanmıştır. Bu problem, çok bilinen bir sentetik gerçek uzay problemiolup, çoklu düzlemsel yüzey üzerinde birkaç tepe oluşurarak günceller. Kıyaslamaprobleminin farklı örnekleri üzerinde yapılan deneysel değerlendirme, geliştirdiğimizyöntemlerin uygulanabilir olduğunu göstermiştir. EFWA tabanlı iyileştirmelerimiz, ilgiliprobleme yönelik göz önüne alınan birçok testte, literatürde yer alan çalışmalardan hemçözüm kalitesi hem de hesaplama maliyeti bakımından daha iyi sonuçlar vermiştir.
Most of the real world problems in different domains demonstrate variouscharacteristics of dynamism, which is one of the major driving force for researchersstudying dynamic optimization problems with an increasing rate for the last two decades.Since changes occur over time in a dynamic optimization problem, the goal of thetarget algorithm becomes tracking the trajectory of the changing optima over time.Evolutionary algorithms and various swarm intelligence techniques have been proposedin the literature to solve dynamic optimization problems. Fireworks Algorithm (FWA) is arecently proposed swarm intelligence algorithm for global optimization of complex staticfunctions that simulates explosion process of fireworks. Although, a set of improvementsover the conventional fireworks algorithm are presented in the literature for the staticoptimization problems, the most evident extension is the Enhanced Fireworks Algorithm(EFWA). In this thesis, three different extensions of the fireworks algorithms are proposedfor solving dynamic optimization problems in real space. The performance evaluationof our EFWA-based algorithms is validated with the Moving Peaks Benchmark, a wellknown synthetic problem in real space that generates and updates a multidimensionallandscape consisting of several peaks. Experimental evaluation on various instances of thebenchmark clearly shows the applicability of our extensions. Our EFWA-based extensionsoutperform the related work in terms of both quality of solutions and computational costfor a large set of test instances of the benchmark.
Most of the real world problems in different domains demonstrate variouscharacteristics of dynamism, which is one of the major driving force for researchersstudying dynamic optimization problems with an increasing rate for the last two decades.Since changes occur over time in a dynamic optimization problem, the goal of thetarget algorithm becomes tracking the trajectory of the changing optima over time.Evolutionary algorithms and various swarm intelligence techniques have been proposedin the literature to solve dynamic optimization problems. Fireworks Algorithm (FWA) is arecently proposed swarm intelligence algorithm for global optimization of complex staticfunctions that simulates explosion process of fireworks. Although, a set of improvementsover the conventional fireworks algorithm are presented in the literature for the staticoptimization problems, the most evident extension is the Enhanced Fireworks Algorithm(EFWA). In this thesis, three different extensions of the fireworks algorithms are proposedfor solving dynamic optimization problems in real space. The performance evaluationof our EFWA-based algorithms is validated with the Moving Peaks Benchmark, a wellknown synthetic problem in real space that generates and updates a multidimensionallandscape consisting of several peaks. Experimental evaluation on various instances of thebenchmark clearly shows the applicability of our extensions. Our EFWA-based extensionsoutperform the related work in terms of both quality of solutions and computational costfor a large set of test instances of the benchmark.