標題: 資料探勘技術應用於晶圓代工需求鏈管理之研究
A Study of Using Data Mining Techniques for Wafer Foundry Demand Chain Management
作者: 鄭稚農
Chih-Nung Cheng
劉敦仁
Duen-Ren Liu
管理學院資訊管理學程
關鍵字: 需求鏈管理;晶圓代工;資料倉儲;資料探勘;決策樹;類神經網路;貝式分類法;Demand Chain Management;Wafer Foundry;Data Warehouse;Data Mining;Decision Tree;Neural Network;Naïve Bayse Classifier
公開日期: 2005
摘要: 有別於傳統的供應鏈管理,需求鏈管理著眼於有效管理客戶需求,並將策略貫徹於生產、配銷等企業流程中。以晶圓代工業為例,其產業特性為提供高度客製化的設計與製造服務,故其需求鏈管理的重要工作,是計算客戶的各項商業指標,給予適當的分類;再針對不同類型的客戶,依其需求進行適當的產能分配。這些決策參數通常被儲存在各種不同類型的系統中,規劃師常耗費大量的時間做資料收集與分析,才能做出客戶的分類規劃以及產能與需求的配置。因此如何設計一個有效率的方法協助做客戶分類以及產能與需求的調配,便成為一項重要的課題。 本研究運用資料倉儲以及資料探勘的技術,建構一套晶圓代工需求鏈管理模型。首先從各資訊系統中收集及計算客戶分類所需的各項指標,儲存於資料倉儲中,做為客戶評分的參考資料;再透過啟發式演算法,提供客戶評分及分類的建議―本研究使用決策樹、類神經網路以及貝式分類法等三種方法進行資料探勘;最後再針對客戶類別,透過本研究提供的演算法,自動進行產能與需求的調配。各客戶分類演算法之準確度、效率以及優劣,將被詳細的分析與比較;而各演算法所探勘出的分類規則以及各決策參數對決策的影響的程度,也在本文中有進一步的討論。
Differentiate from traditional supply chain management; demand chain management focuses on managing customer demand efficiently and execute the strategic instructions in production and distribution flows. In wafer foundry industry, manufacturers provide high-customized design and manufacturing services to their clients. The most important work of demand chain is to collect the business indices of customers, to classify them into several categories using the indices and allocate proper capacity volume to customers according to their categories and demand. But the decision parameters are stored in various information systems, planners always spend much time to collect and analyze data in order to plan the customer classification, capacity allocation and demand allocation. It would be an important issue to design an efficient method to do the classification and allocation. This paper constructs a prototype of wafer foundry demand chain management using data-warehouse and data-mining techniques. First, it extracts the decision indices from various systems and loads them into a data warehouse. The indices would be an important reference of customer scoring. Some heuristic algorithms are used to calculate the indices and provide the recommendation of customer scoring and classification ― decision tree, neural network and Naïve Bayse classifier are used in this study. According to the customer categories, some algorithms in this study are used to allocate capacity and demand properly. The accuracy, performance, advantages and disadvantages will be analyzed and compared. The parameter relations that are found by the methods will be discussed in this study.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT009364502
http://hdl.handle.net/11536/79987
顯示於類別:畢業論文