Genetic Algorithms with Fine-Tuning Mutation on Elite for Multi-Modal Function Optimization

• Main Authors: LU, SHU-HAO, 盧書豪
• Other Authors: 呂森林
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/01355974655115941452

碩士 === 國立臺灣科技大學 === 機械工程系 === 93 === This thesis proposes an efficient approach for multimodal function optimization using Genetic Algorithms (GAs). We recommend the use of Fine-Tuning Mutation with Elitist Strategy (FTME) to realize the twin goals of perfecting the existing solution (exploitation) and maintaining the searching capability to the whole solution space (exploration). To be specific, the accuracy is enhanced without losing the genetic diversity of population. In the GA using FTME (GA-FTME), a novel genetic operator called fine-tuning mutation is proposed. The device of fine-tuning mutation and the combination mechanism of FTME are interpreted. The performances of the GA-FTME and the Standard GA (SGA) with elitist strategy are compared in optimizing several massively multimodal functions with varying complexities. For most functions, the GA-FTME converges to the global optimum precisely in fewer generations than the SGA, and it achieves a higher convergent probability. The average rate of improvement on convergent probability is 24.43(%). The average rate of improvement on the average critical generations is 14.50(%). The average rate of improvement on the average computation time is 12.81(%). The simulations also demonstrate that the FTME can be applied to most settings of parameters in GAs, such as using different population size and different maximum generation number. We believe that the FTME has a high applicability to most of GAs.