以直交表为基础之智慧型演算法AnIntelligentAlgorithmBasedon.PDF

吳鳳學報 第二十期 以直交表為基礎之智慧型演算法 An Intelligent Algorithm Based on Orthogonal Array 1 2 何榮松 ，楊淵成 1 機械工程系 吳鳳科技大學 2 機械工程系 吳鳳科技大學 摘要 近二十年來，實驗計劃法已為大家所接受並應用在許多工程問題。由於其利用直交 表設計以少量的實驗次數獲得高效率的情報特性，因此直交表常作為演算法之一部份以 提高各演算法的計算效率及可靠度。 本文提出新穎的 IOA-DOE 。具有以下特性:一. 適合於複雜的非線性問題，二. 適 合於求導數非常困難或甚至無法求得導數之問題，三 . 演算法之收斂速度，並不受限於 問題維度 (Dimension)之大小，四. 有效結合直交表獲得的梯度資訊及直交表之高情報 效率，使得解的收斂速度及準確度獲得進一步提昇， 五. 求解速度及精度不受初始解 的選擇而影響，及六 . 突破梯度法這類演算法之限制，亦即，皆能搜尋到全域之極值。 本方法之初始解可以亂數方式隨機產生，求解的速度及精度幾乎不受影響。 關鍵詞 :梯度法、直交表、最佳化問題 Abstract In last two decades, owing to the excellent performances, e.g., extracting useful information from small amount of experiments and provides an efficient collection of most appropriate design parameters, orthogonal array analysis has widely applied to versatile engineering problems successfully. Therefore, some well-known evolutionary algorithms always have been merged with orthogonal array analysis in order to improve their own poor efficiencies. In this paper, lots of well-known benchmark functions were utilized to verify the performances of the proposed approach, IOA-DOE (Intelligent Optimization Algorithm based on Design of Experiments). Design parameters collections obtained from orthogonal array experiments are efficient in specific searching space. The proposed method has the merits of both the local search method and global search method. It is noted that some drawbacks of orthogonal experiment design were discussed and conquered. From some test results in this paper, the proposed method h