標題: LED 封裝體需求預測-以智慧型手機產業為例-
Forecasting Light-Emitting Diode Package For the Smart Phone Industry
作者: 李宗翰
姜齊
LI , Tsung-Han
管理學院管理科學學程
關鍵字: 發光二極體;需求預測;時間序列;Light-emitting diodes;Forecasting;Time series
公開日期: 2016
摘要: 台灣為一出口導向的國家,其中出口佔國家GDP約70%,電子製造業又佔了出口的30%。需求預測在各行各業都是少不了,從日常生活的蔬菜需求到整體國家經濟預測,從企業財務規劃、庫存管理、生產計劃、到客戶管理等各層面,需求預測都是一個最關鍵也是最基本的數字。這個數字是如何產生,大家都只有一個希望,那就是愈準確愈好。不準確的需求預測會帶來什麼樣的結果呢?當預測大於實際需求會造成過多的庫存、庫存成本資金積壓、產生轉運成本增高、報廢庫存增高、降低獲益率。當預測小於實際需求,會造成生產排程效率降低、較高的產品成本、錯失銷售時機、客戶滿意度降低。因此如何提高需求預測的正確性,成為企業經營中一個重要的課題。
要如何降低需求預測的不確定性呢?對於具備歷史資料的產品來說,利用過去的資料進行預測是最簡便的方法,然而因為LED處於產業上游位置,所以需要用一些簡單的轉換關係,就可以產生客觀的預測結果。本研究應用不同的預測模式,線性迴歸、非線性迴歸、ARIMA以及SARIMA來探討智慧型手機LED封裝體未來需求量。利用2009年1月至2014年12月智慧型手機出貨量為樣本內資料,藉由不同的預測模型來進行樣本內的預測,並將此需求預測模型所產生之2015年預測結果與該年實際智慧型手機進行比較,最後再進行LED需求轉換。這種做法,可以減少長鞭效應的產生。結果顯示,本研究所使用之非線性迴歸預測模型MAPE值最佳,得到最佳化的LED需求預測。整體而言也會比使用線性迴歸、ARIMA之需求預測績效來得好,增進約7~9倍。
要百分之百準確的預估未來幾乎是不可能的事,但我們仍可以藉由增加不同參數變化,來改變需求預測模型。本論文希望透過相關之需求預測模型,改變LED需求預測的方式,做為產業界在從事需求規劃時的一個參考,進而能提昇LED產業之需求
測品質,增進供應鏈管理之效率。
Taiwan is an export-oriented country, with exports accounting for about 70% of the GDP. The electronics manufacturing industry has accounted for 30 percent of exports. Forecast has only one question, how to be more accurate and better. If inaccurate forecasting what will bring results? When the forecast is greater than actual demand will result in excess inventory, inventory backlog of capital costs, resulting in higher transport costs, increased inventory scrap and reduce the benefit rate. When less than the actual demand forecasting, production scheduling will cause reduced efficiency, higher product costs and missed sales opportunity, customer satisfaction decreases. So how to improve demand forecasting accuracy? This has become an important problem.
How to reduce the uncertainty of demand forecasting? In this study, different forecasting models, linear regression, nonlinear regression, ARIMA and SARIMA to discuss the future demand for smartphones LED package. Use from January 2009 to December 2014 smartphone shipments in the sample data, with different forecasting models to predict within the sample, and this demand forecasting model arising in 2015 and predictions. This approach can reduce the production of bullwhip effect. The results show that the nonlinear regression model has the best MAPE value than other demand forecast model. Overall, the ratio will be using linear regression it can increase of about 7 to 9 times than linear regression or ARIMA.
To be 100% accurate estimate of the future is almost impossible, but we can still increase by different parameters, to change the demand forecasting model. The paper hoped that through the relevant demand forecasting model, change the LED demand in a predictable manner which in turn can enhance the needs of the LED industry measuring quality, enhance the efficiency of supply chain management.
URI: http://etd.lib.nctu.edu.tw/cdrfb3/record/nctu/#GT070363126
http://hdl.handle.net/11536/138675
Appears in Collections:Thesis