| Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 99 === Genetic algorithm is widely used to solve the optimization problem. The success of GA can attribute to the tight building BBs which are generated in the evolutionary process, but tight building BBs are not always formed for all problems, especially on GA-hard problems. Therefore, our goal is to find the linkage between variables and bind them to form BBs. This study proposed the method that combines GA, similarity measure and clustering algorithm for linkage identification. The genes with dependency would be bound together to form BBs. The clustering algorithm is used to group genes according to similarity measure. Each group is taken as a BB. This research adopts affinity propagation (AP) to deal with clustering problem. The input information, for the AP, comes from similarity measure. Then GA evolves the better solution and BBs simultaneously.
Experimental results show that evolving BBs and solutions together can reduce the computational costs. However, it uses more population. Separately identifying BBs and exploring solutions will confront with a problem which is how to determine that the accuracy of BBs, without any given information, is enough to enter into the next step. As a whole, evolving BBs and solutions together has good performance and saves the parameter setting issue.
|
|---|
