本文目的在探討基因演算法在矩陣實驗室的單一程式/多工處理之實作設計與效能分析，我們發現在使用矩陣實驗室實現平行運算時，可能會造成一些問題，並提出解決方案。我們以旅行者問題為基礎設計程式，採用平行運算改良程式，將基因演算法程式迴圈平行化，初始族群分割到多個工作區域同時運算，以提高程式運算速度。此外，此方法也使得程式不易收斂到局部最佳解，所產生平行路徑可以對初始族群做重複修正。當初始族群總數和演化代數總數相同時，我們發現平行基因演算法具有更高效能並找到更優化的解答。 In this thesis, we aim to design and analyze the genetic algorithm implemented on single program/multiple data for Matlab. Some experiments based on the Travelling Salesman Problem were conducted and parallel program were developed for it. For-loop programs are paralleled, initial population was split to multiple work area for speed-up computation. Additional advantages were such as local optimization could be avoided, and initial population can be adjusted repeatedly by using each parallel path generated. When the number of initial population and the generation number of the final convergence are the same, we found that the parallel genetic algorithm has higher performance and better solution.
