在記憶體產業中，由於產品市場競爭環境激烈及變動快速，營運策略需兼顧成本、品質、速度與彈性。這些面向往往互相影響，例如:若要滿足訂單變動的達交速度，需準備一定水位的安全庫存以滿足客戶的期望交期，在此同時，庫存成本與呆滯風險也將隨之上升。因此，如何取得及運用商業資料數據分析輔助企業製作營運計畫與決策，成為重要的課題。
一般企業在備料採購數量估算會採用統計迴歸或MRP系統工具。隨著電腦科技進步與人工智慧興起，類神經網路在各領域應用日見效益，因此本研究首先建議運用「類神經網路」建立備料模型，並與傳統的迴歸統計模型及MRP備料模式共同評量，藉由最佳備料模型來推測物料採購量。接著，並模擬多個安全庫存水位方案，分析及量化營運指標，包含達交率、風險庫存與商業利益，協助管理者提升商業決策品質。
經由個案研究發現「類神經網路」預測能力最高。無論營運策略著重達交率或風險庫存，「類神經網路模型」績效表現都是最優。因此，安全庫存水位方案可運用「類神經網路模型」模擬，因其嚴謹度與適切度較迴歸統計模型或MRP工具都高，可提供較高參考性的量化數據給營運主管，以優化備料決策。本研究可供各企業參考，解決「及時達交率」與「降低風險庫存」兩難的營運策略決策課題。

In the memory industry, the market is undergoing intensive competition and rapid change. The enterprise’s operating strategy has to consider cost, quality, time and flexibility. These aspects often affect each other. For example, in order to meet the customer's expected delivery time which changes often, it is necessary to prepare safety stock. At the same time, stock cost and risk inventory will increase. Therefore, how to select the right approach to assist the enterprise in making appropriate decisions of safety stock becomes an important issue.
Because of the advancing computer technology and the emergence of Artificial Intelligence, the application of neural network is increasingly becoming popular in various fields. In response to this new application, this study seeks to: (1) use neural network to build the material forecast model, (2) use the model which has best forecast capability estimate materials requirement, and (3) simulate multiple safety stock proposal, analyze and quantify operational indicators, including on-time delivery rate, risk inventory and sales margin, to help the managers enhance the quality of decision-making.
Through analysis of the case study, it is found that neural network model has the highest predictive ability. No matter the operation strategy focuses on on-time delivery rate or lower risk inventory, the performance of the neural network is the best. Therefore, the safety stock proposal can be simulated using the neural network model to enhance decision-making quality. This study provides a useful reference for the enterprise to combat the challenge of operating strategy decision-making between on-time delivery rate and lower risk inventory.

第一章 緒論………………………………………………………1
1.1 研究動機…………………………………………………………1
1.2 研究目的…………………………………………………………2
1.3 研究流程…………………………………………………………2
第二章 相關文獻探討…………………………………….4
2.1 營運策略與指標……………………………………….…4
2.2 備料與庫存管理………………………………………….…4
2.3 資料探勘……………………………………………………………5
2.4 迴歸分析……………………………………………………………6
2.5 類神經網路………………………………………………….…9
2.6 MRP 系統…………………………………………………………11
2.7 相關研究文獻探討…………………………………………11
第三章 研究方法………………………………………………14
3.1 研究架構……………………………………………………………14
3.2 資料準備……………………………………………………………15
3.3 模型發展與建立………………………………………………15
3.3.1 類神經網路……………………………………………………15
3.3.2 迴歸統計分析………………………………………………16
3.4 運用最佳模型與備料決策………………………………16
第四章 個案研究………………………………………….……18
4.1 個案公司背景說明……………………………………………18
4.2 資料收集與影響因子………………………….…………20
4.2.1 資料收集…………………………………………………………20
4.2.2 影響因子………………………………………………………20
4.2.3 資料整理與篩選……………………………………………21
4.3 建立最佳備料模型……………………………………………22
4.3.1 類神經網路模型建立…………………………………22
4.3.2 多元迴歸模型建立………………………………………24
4.4 備料模型預測績效比較………………………….….…26
4.5 營運目標與安庫決策………………………………………27
第五章 結論與建議………………………………………………32
參考文獻 ……………………………………………………………….…34
1 中文 ………………………………………………………………..……34
2 英文 …………………………………………………………………..…35

1 中文
[1]蘇朝墩 (2013)，品質工程:線外方法與應用，前程文化事業有限公司。
[2]蘇朝墩 (2019)，Introduction to Deep Learning上課教材。
[3]林聰明、吳水丕 (1981)，指數平滑法之選擇與應用，華泰書局，台北。
[4]宋鎧、范錚強、紀延平、郭鴻志、陳明德(1993)，管理資訊系統，國立空中大學。
[5]春日井博 (1988)，需求預測入門，書泉出版社，台北。
[6]https://www.inside.com.tw/article/9677-dram-industry
[7]桑慧敏 (2018)，一生受用的統計學，大數據分析之鑰。
[8]產業價值鏈資訊平台網站，台灣證卷交易所。
[9]何應欽 編譯(2018)，作業管理，第十三版，William J. Stevenson著。
 
2 英文
[1]Boldt, B. I. (1982). Sound business forecasting. Today's Executive, 5(1), 6-11.
[2]Eswaran, C., & Logeswaran, R. (2010, May). A comparison of ARIMA, neural network and linear regression models for the prediction of infant mortality rate. In 2010 Fourth Asia International Conference on Mathematical/Analytical Modelling and Computer Simulation (pp. 34-39). IEEE.
[3]Han, J., Pei, J., & Kamber, M. (2011). Data mining: concepts and techniques. Elsevier.
[4]Hung, Y. F., & Chang, C. B. (1999). Determining safety stocks for production planning in uncertain manufacturing. International Journal of Production Economics, 58(2), 199-208.
[5]Law, R., & Au, N. (1999). A neural network model to forecast Japanese demand for travel to Hong Kong. Tourism Management, 20(1), 89-97.
[6]Létourneau, S., Famili, F., & Matwin, S. (1999). Data mining to predict aircraft component replacement. IEEE Intelligent Systems and their Applications, 14(6), 59-66.
[7]Makridakis, S., & Wheelwright, S. C. (1977). Forecasting: issues & challenges for marketing management. Journal of Marketing, 41(4), 24-38.
[8]Reid, R. D., & Sanders, N. R. (2012). Operations management: an integrated approach. John Wiley & Sons Incorporated.
[9]Sohl, J. E., & Venkatachalam, A. R. (1995). A neural network approach to forecasting model selection. Information & Management, 29(6), 297-303.
[10]Stevenson, W. J. (2020). Operations Management. McGraw Hill.
[11]Yao, J., Teng, N., Poh, H. L., & Tan, C. L. (1998). Forecasting and analysis of marketing data using neural networks. J. Inf. Sci. Eng., 14(4), 843-862.