| Summary: | 碩士 === 國立暨南國際大學 === 資訊管理學系 === 103 === Global optimization means finding the global optimal solution in the enormous solution space. Due to different problem natures, the fitness landscape varies. Some problems have prickly landscapes some others have smooth ones. According to the No-free-lunch theorem, there exists no such an algorithm that can fit to all kinds of problems, leaving global optimization still a challenging problem. This study proposes a tripartite evolutionary architecture, under which an Opposition-Based Co-evolution Learning Tabu Search (OCLTS) scheme is presented. The OCLTS is majorly based on tabu search and uses landscape analysis to analyze the profile of executed course, so as to predict the landscape of forthcoming course and to adapt the step size of the solution movement. Our method employs the restart mechanism in order to avoid getting trapped by local optimality. The adoption of the opposition point to the incumbent solution is useful in increasing the population diversity. Moreover, the co-evolution scheme is deployed to exchange information between populations such that the premature convergence can be prevented. The performance of the proposed tripartite evolutionary architecture is validated by a number of benchmark global optimization functions. The results show that our method obtains good quality solutions for most of the benchmark functions.
|
|---|
