| Summary: | 博士 === 國立中興大學 === 應用數學系所 === 94 === In this dissertation, new methodologies were proposed to design evolutionary algorithms. More specifically, two-phase and iterated two-phase evolutionary algorithms were proposed in order to arrange properly the strength of intensification and the strength of diversification. In the first phase, diversification is enhanced that aims to search broadly for promising areas. In the second phase, intensification is enhanced that aims to search more thoroughly within each promising area.
Two resource-constrained project scheduling problems (RCPSP) was considered to test our new methodologies. We first proposed a metaheuristics named ANGEL which combines the ant colony optimization, the genetic algorithm and the local search method for the single mode RCPSP. The second algorithm proposed is a two-phase and a iterated two-phase genetic local search algorithms for the multi-mode RCPSP. By suitable application of the evolutionary operators, the local search method, the restart of the algorithm and the collection of good solutions in the elite set, the strength of intensification and diversification can be properly adapted and the search ability can be retained even in a very long term.
A series of experiments were conducted to assess the capabilities of these evolutionary algorithms. Instances of these problems were taken from well known benchmarks with different problem sizes.
|
|---|
