Development of hybrid methods based on bat algorithm for large-scale continuous optimization problems

Bat Algorithm (BA) is a nature-inspired metaheuristic algorithm proposed in 2010. The Bat algorithm is an algorithm that experiences performance decreases due to dimension increase, as in other metaheuristic algorithms. In this thesis, two hybrid algorithms have been proposed to reduce the structural problems of BA and increase its performance on large-scale problems. The first hybrid algorithm (MBADE) has been created by using BA and Differential Evolution (DE) algorithms together to contribute to the local search capability of BA. In this algorithm, the algorithm to be applied to the individual in each iteration is decided according to a probability value based on performance. The second hybrid algorithm (BA_ABC) has been developed by using BA and Artificial Bee Colony (ABC) algorithms together to increase the global search capability of BA. In this hybrid algorithm, the population is divided into two subpopulations, and BA and ABC algorithms run on different subpopulations. When certain conditions are provided, information exchange is made between subpopulations. The proposed hybrid algorithms have been tested on CEC2005 small-scale benchmark functions, CEC2010 large-scale benchmark functions, and large-scale problems selected from CEC2011 real-world problems. The obtained results have been compared with the results of both BA versions and other metaheuristic algorithms selected from the literature. Besides, the results have been interpreted with the help of statistical tests, and it has been examined whether there is a significant difference between the algorithms. The proposed hybrid algorithms have produced better results than the standard BA algorithm for most of the tested benchmark functions. While the MBADE algorithm is more successful in small-scale benchmark functions, the BA_ABC algorithm is more successful in large-scale benchmark functions. Comparisons with the BA versions and other algorithms selected from the literature show that the proposed hybrid algorithms have produced successful, competitive, and acceptable results.