| Summary: | 博士 === 國立高雄第一科技大學 === 工學院工程科技博士班 === 105 === In this present study, Taguchi and uniform design (UD) are embedded in genetic algorithm (GA); then attempted in the three cases are discrete and continuous problem; this procedure aims for improving its searching performance.
In the first problem, a two-levels of Taguchi’s table is embedded in GA between crossover and mutation named Hybrid Taguchi Genetic Algorithm (HTGA). This algorithm outperforms previous research for optimizing PCB drilling path. Meanwhile, UD is used as the design experiment for preparing the training data in a neural network (UniNeuro), then forming the metamodel of unknown system model in the second case. UD is also used in initialization and enrichment procedure after mutation in GA to obtaining several input parameter settings for multi objectives (HUDMOGA). This algorithm, UniNeuro-HUDMOGA was successfully received several optimal input parameter settings in double inverted pendulum. In the third case, for optimizing Laser-Auto-Focus-Based Tracking System (LAFS) the similar design with the second case is employed. However, in this case single objective is required, called hybrid uniform design genetic algorithm (HUDGA). In this case, UniNeuro-HUDGA is used for minimizing the tracking error by LAFS on scanning the reference profile - and then outperforming the trail-error by expert person.
As the result, the research proved that hybridizing the Taguchi table and UD in the GA adequate improve the searching achievement; and prevent trap in local optima. UD is efficient to apply because its flexibility in parameter and experiment number, even number of levels. The metamodeling that improved by UD could be used as the vast useful method for generating the fitness function when the true function is undefined. Moreover, UD has more chance to generate the possible solution accordingly the uniformly spreading candidate on whole search space. Henceforth, the sustainability of UD research is required for gaining the optimization design.
|
|---|
