Evolutionary algorithms and simulated annealing are the most frequently used stochastic optimization algorithms. The evolutionary algorithms modeled their mechanism on the natural evolution. The main concept of the evolutionary algorithms is competition among individuals. On the other hand, the simulated annealing imitated the annealing process of an object. A proper annealing gives the object the crystalline structure, the lowest energy state.
The subject of this thesis is to propose more powerful optimization algorithm through unification of these two famous stochastic optimization method. The basic structure of the proposed algorithm follows the simulated annealing, while the selection is controlled by competition among individuals.
The proposed algorithm has two merits. By using temperature at the generation probability density function, the balance between exploration and exploitation can be effectively controlled and as a result, we can avoid the premature convergence which have long been tackled the evolutionary algorithms. Also, by inducing competition, we can make the algorithm to converge more quickly than the simulated annealing.
The proposed method also can be applied to the multiobjective optimization. We prove that after a number of iterations, the random selection makes some of the found nondominated solutions to disappear and converge to a uniform one. Therefore, we introduce elitism in our algorithm as a deterministic selection rule. The repulsive mutation strategy is additionally proposed for controlling the excessively exploitive selection rule.
We completely prove the convergence of the proposed algorithm. The performance of the new algorithm is compared with the existing optimization method using test problems with various characteristics. At the experiments, the proposed algorithm shows better performance than the other algorithms. Moreover, we proposed a method for designing optimal analog filter as a real-world application of the propose...