New approaches to solving large-scale optimization problems

Optimization is used in real-world problems to speed up decision-making processes or to improve decision-making quality. Problems with many parameters that need to be optimized are called large-scale global optimization (LSGO) problems in the literature. As the number of decision variables increases, the search space and thus complexity of the problem increases exponentially. For this reason, it is necessary to design strong algorithms and methods to overcome the problems that may arise, or to develop existing algorithms by supporting them with different mechanisms. Within the scope of this thesis, non-decomposition based solution approaches that deal with the problem as a whole have been studied for the solution of LSGO problems. As part of the development of standard evolutionary algorithms with additional editing and techniques, changes were made in the helical motion stage of the Artificial Algae Algorithm (AAA), which contributes to both the exploration and exploitation mechanisms, and the developed version of the algorithm was named Modified Artificial Algae Algorithm (MAAA). The tests performed with the CEC2010 test function set showed that the change made increased the LSGO solution performance of the algorithm. Another effective solution approach preferred to cope with LSGO difficulties is memetic algorithms. Within the scope of this thesis, a new local search method called Golden Ratio Guided Local Search with dynamic step size (GRGLS) has been developed to be used in the local search phase, which has a very important effect on the performance of memetic algorithms. The tests performed with the CEC2013 test function set have proven that the proposed algorithm achieves the best solutions among the compared algorithms and outperforms the others in overlapping functions and non-separable functions. To further develop the proposed GRGLS method, the use of chaotic maps in the field of optimization has been investigated. The non-repeating random number generated from the Singer chaotic map is used as a coefficient in the step size determination equation that provides movement in the search space. The new version developed is called Chaotic Golden Ratio Guided Local Search (CGRGLS). Three separate performance evaluations with the CEC2015 Big-Opt test function set confirmed that the new version of the local search method achieved better results than the previous version. All the results obtained from the studies have shown that the proposed GRGLS and CGRGLS local search methods are effective and efficient local search methods that can be used in the LSGO area.