| Summary: | 博士 === 國立臺灣科技大學 === 機械工程系 === 92 === Genetic Algorithms have been applied on many optimization problems with success and the most common strategy to handle constraints is penalty strategies. The success of the genetic search highly depends on appropriate selections of many operation parameters, especially the parameters associated to penalty strategies, which are often determined through the trial-and-error process or by experience. Furthermore, the gene loss problem often existing in the binary-coded genetic algorithms (BGA) will become worse when the penalty strategy is involved. This dissertation aims to devise an adaptive and parameter free penalty strategy: the first- and second-generation self-organizing adaptive penalty strategy (SOAPS & SOAPS-II) to avoid the agonizing selection of penalty parameters and to increase the reliabilities of genetic searches as well. In this work, a hybrid-coded crossover strategy (HCC) is also proposed to increare the effectiveness of attainment of the global optima for the genetic searches with small population sizes. By combining the advantages of crossover strategies of real-coded genetic algorithms (RGA), HCC can significantly reduce the gene loss phenomenon in BGA and make the BGA search robustly and reduce the sensitive influence of parameter selection. All combinations of proposed strategies are tested and compared with other combinations of known penalty strategies in mathematical and engineering optimization problems with favorable results. The test results also show the combinations of HCC and SOAPS, and HCC and SOAPS-II consistently outperform other strategy combinations.
|
|---|
