| 摘要: | 進化式演算法已經被廣泛的應用在許多領域，例如高能物理分析、氣象預測、基因分析、生物研究、財務或商業資訊分析等。雖然這些方法能夠提供令人滿意的解決方案。然而不幸的是他們都已經被證實為是耗時的方法。因此,大量運算能力的需求隨之出現，雖然雲端的出現大幅的提升了運算的效能，但同時也提升了問題的複雜度，例如:更多的資料維度或資料數。此外，基於No Free Lunch (NFL)理論，所有的進化式演算法都必須付出相同的代價進行求解，換句話說，若要得到更好的解，相對的必須付出更多的代價，例如更多的作業，流程的改變，更多的時間，更多的個體等。
因此本研究目的在於提供一個以Divide and Conquer (D&C)的想法為基礎，實作一個架構以適用於所有進化式演算法克服搜尋空間過大的問題(如:龐大的搜尋空間，過早收斂，資料篩檢，黑箱等)，並能夠協助進化式演算法紀錄搜尋過程中每個維度的敏感度資料，幫助使用者能夠更深入了解問題所在。
;Evolutionary algorithms have been widely used in many fields, such as high energy physics analysis, weather forecasting, genetic analysis, biological research, financial or business information analysis. Although these methods can provide some satisfactory solutions, they have been proved to be time-consuming methods. Therefore, the computational efficiency needs to be taken into account. Although the cloud technique can significantly improve the computing performance, it also increase the complexity of problems, such as more dimensions or numbers of the datasets. In addition, based on No Free Lunch theory (NFL), all evolutionary algorithms must pay the same price for solving problems. In other words, to get a better solution, it must pays more costs, such as more operations, process change, time or individuals.
Therefore, this thesis aims to provide a framework based on the idea which is according to the Divide and Conquer (D&C) principle, and this framework can be applied to all evolutionary algorithms to overcome the large search space problem that can affect the computational efficiency (such as: a huge search space and premature convergence, data screening, black box, etc.). In addition, it also can assist the evolutionary algorithms in recording the sensitive data samples in each dimension during the searching process, which helps users to fully understand the problem. |
