| Summary: | 碩士 === 國立中山大學 === 資訊工程學系研究所 === 103 === In the past decades, many global optimization algorithms based on biologically-inspired strategies have been developed. Most of them are population-based algorithms and their abilities of adaptive learning have shown they can solve optimization problems effectively. Particle swarm optimization (PSO) is a very popular and common-used strategy among them since it is easily implemented. In the early days, PSO was usually performed on a single swarm. Along with the development of variant PSO technologies, multiple-swarm schemes were also adopted for some purposes such as parallel processing, multimodal optimization and multi-objective optimization. In the thesis, we first revisit and discuss some interesting characteristics of PSO for multiple sub-swarm processing. We then propose a multi-sub-swarm algorithm, in which the original particle swarm is divided into several sub-swarms with the same total size, to investigate the variation of performance with different sub-swarm numbers. The algorithm is very suitable to be parallelized in nature. We then propose a hierarchical PSO strategy called HPSO, which executes the PSO algorithm in hierarchical levels and uses some operations to increase the performance. Different execution structures of HPSO are discussed as well. Furthermore, we propose a dynamic migration mechanism for PSO, which can automatically determine when to migrate a portion of particles from one sub-swarm to its neighbor. Finally, we apply the dynamic migration mechanism on the HPSO to check the effects of combination. By additionally using some operations such as migration, merge and re-initialization, the particles can increase diversity effectively and thus obtain good results. Experiments are also made to show the performance of the proposed approaches.
|
|---|
