| Summary: | 博士 === 國立東華大學 === 電機工程學系 === 95 === This dissertation proposes numbers of variants of population-based optimizers. Several solution searching strategies and mechanisms are proposed to improve solution searching capability and efficiency of population-based optimizers for dealing with different types of optimization problems. To verify solution searching ability of proposed algorithms in solution space, the proposed algorithm will be applied for solving numerical optimization, blind source separation and placement constraints for VLSI design. For numerical optimization, this dissertation proposes a sifting strategy for population called population manager and sharing principles, and applies them to particle swarm optimizer (PSO) to enhance its solution searching ability for high dimensional numerical optimization. In the experiments, 15 test functions are selected for comparison among proposed method and recently related works. From the results, the proposed method can effectively jumping out of local optimal, showing rapid convergence, saving computation time and is able to find out near or real global optimal solution. Besides, genetic algorithm (GA) is also a kind of population-based optimizer; the GA utilizes the evolution of chromosomes for finding optimal solution. In this dissertation, the strategies that mentioned above are also applied to GA. In the experiments, 18 test functions which selected from CEC 2005 technical report are utilized for comparison among proposed method and recently related works. The improved genetic algorithm is also with well ability for jumping out from local optimal, rapid convergence, and be able to find out near or real global optimal solution. For real-world applications, this dissertation improves particles’ searching behavior for PSO to fit different solution searching characteristic of applications. The improved PSO is applied to learning rate adjustment for blind source separation (BSS) problem, which can separate each source signal from mixed signal efficiently and completely. Besides, the improved PSO is also applied to placement constraints for VLSI design, for finding optimal placement solution with minimized chip size and wire length. Comparing with algorithms proposed by other recently related works, the proposed methods exhibit better results for efficient solving both applications.
|
|---|
