應用粒子群最佳化演算法於多目標存貨分類之研究 研究生:葉思緯 指導教授:蔡啟揚 教授 元智大學工業工程與管理研究所 中文摘要 在成千上萬的存貨種類中,包含著各種性質差異極大的物項,想施以完全相同的管理方法來處理所有的物項,並非一個聰明的做法,而最好的方法則是以存貨管理的「目標」為依據,將存貨物項分類為數個不同的群組,最後再針對個別不同的分群採取不同的管理手法加以控管。 本研究將粒子群最佳化演算法應用於存貨物項之分群,在不需事先告知所要分群數目的情形下,將存貨物項自動分群為最佳之分群數目,並使分群結果能夠滿足不同目標之目標式,包括成本目標式、需求關聯性目標式、存貨週轉率目標式,另外整合上述三個目標式成為多目標目標式。最後將業界實際的存貨物項資料代入粒子群最佳化分群法,並與一般常用的分群法則做比較,如供應商分群法、ABC分析法及將物項歸類為一群等,經由實驗設計與結果分析顯示,粒子群最佳化分群法均可以獲得最佳的分群結果。除上述的結果之外,本研究也探討了粒子群最佳化演算法的參數設計,對於粒子數目、搜尋次數、學習因子、最大粒子速度及慣性權重等容易影響搜尋結果之參數作一分析與討論,最後並提出參數設定之建議值。 關鍵字:粒子群最佳化演算法、多目標、存貨分類
Multi-objective inventory classification using Particle Swarm Optimization Graduate:Szu-Wei Yeh Advisor:Dr. Chi-Yang Tsai Department of Industrial Engineering and Management Yuan-Ze Unviersity Abstract In thousands upon thousands kinds of inventories, there are different items with various characters. It is not clever to handle all items with the same method of management. The best approach is classifying the items of inventory into several groups in accordance with goal of inventory management, and different groups are controlled in light of different management methods. This research develops an inventory classification method based on Particle Swarm Optimization, and this approach can automatically classify the items of inventory to optimal number of groups. The result of classification satisfies different objective functions, including cost, demand correlations, inventory turnover, and in addition, the research integrates three objective functions to a multi-objective function. Practical data is used to test the performance of the proposed approach. Furthermore, it is compared to those of some general applied inventory classification methods, such as grouping policy according to provider, ABC classification system and all items classified in one group. Through experimental design and result analysis, it is shown that this approach performs the best among all tested methods. In addition, this research also investigates the best selection of parameter values in the Particle Swarm Optimization. The parameters which are analyzed include population size, iteration number, learning factors, max particle velocity and inertia weight. Finally, the proposed approach presents suggestions for setting the values of parameters. Keywords:Particle Swarm Optimization、Multi-objective、inventory classification