倒傳遞類神經網路在銷售預測之應用 以TFT-LCD產業為例

An Application of Back-Propagation Neural Network in Sales Forecasting: A Case Study of a TFT-LCD company

徐品勝(Pin-Sheng Hsu)

指導教授 ： 張百棧

元智大學/工程學院/工業工程與管理學系/碩士(2007年)

摘要

TFT-LCD ( Thin Film Transistor-Liquid Crystal Display；TFT-LCD)產業為我國目前主要發展之新興產業，由於產業的供需快速變化，產品銷售易受到環境因素之影響，因而時常造成供需失衡的現象，目前國內未有許多相關研究論文在TFT-LCD 產業之銷售預測上，因此透過一套準確的銷售預測模型，將可提供未來銷售需求之評估，進而達成生產銷售之供需平衡，以減少成品庫存的成本堆積。
本研究以國內知名TFT-LCD製造廠為例，其研究共分為三大部份；首先為蒐集總體經濟指標、工業生產指標與下游銷售指標等九項因子，輸入倒傳遞網路以此為模式（一）；第二部份為運用逐步迴歸關鍵因子分析，選取對於銷售預測影響性較高的因子，並加入時間序列因子作考量，採取溫氏指數平滑法來衡量預測目標之季節性與趨勢性因子的影響性，並以此為模式（二）；第三部份是以模式（二）為最適模型樣本，再以修正因子預測法進行模式演化，使得網路得以具有完整之誤差回饋機制，而以此為模式（三）；最後，將上述之模式建立成三預測模型，並進行平均絕對百分比誤差法（MAPE）、平均絕對偏差（MAD）與誤差均方根差（RMSE）之衡量指標比較。經由本研究結果得知，倒傳遞網路模式（三）-逐步、溫氏與修正之混合模型，其相較於其他模式有較佳的預測效果，因此，建議樣本公司依據此模型進行銷售預測之評估。

關鍵字

TFT-LCD產業 ； 銷售預測 ； 逐步迴歸分析 ； 溫氏指數平滑法 ； 修正因子預測法 ； 倒傳遞類神經網路

並列摘要

TFT-LCD ( Thin Film Transistor-Liquid Crystal Display；TFT-LCD) is a newly developed industry in Taiwan. Owing to the rapid changes in the supply-and-demand environment, there have been many studies conducted on the sales forecasting to solve the unbalance of supply and demand. Through establishing a precise sales forecasting model, the upcoming demand can be evaluated in advance to achieve the production balance and decrease the inventory cost.
The research is divided into three major phases based on the data collected from a famous TFT-LCD manufacturer in Taiwan: (1) 9 indices are compiled and put into the Back-Propagation Neural Network model. (2) Stepwise Regression Analysis is utilized to sort out factors with higher prediction impact and then Winter’s Exponential Smoothing method is adopted for seasonal time series data analysis to set up a new model. (3) To evolve and equip the optimal model generated from phase 2 with a complete error feedback mechanism. Finally, MAPE, MAD and RMSE are used to compare the index effectiveness of these three models. The results indicate that the hybrid model has the best prediction ability compared to the other two models; hence, it is recommended for business to apply in sales forecasting.

並列關鍵字

TFT-LCD Industry ； Sales Forecasting ； Stepwise Regression Analysis ； Winter’s Exponential Smoothing ； Revise Factor Forecasting ； Back-Propagation Neural Network

參考文獻

38. 陳振嘉，「以類神經網路建構階段式企業財務危機預警模型」，中原大學資訊管理系，碩士論文，2005。

1. Al-Saba, T., I. El-Amin,“Artificial neural network as applied to long-term demand forecasting”.,Artificial Intelligence in Engineering 13 pp.189-197,1999.

2. Gardner,E.S &D.G. Dannenbring, “Forecasting with Exponential Smoothing: Some Guidelines for Model Selection.” Decision Sciences,Vol.11,NO.2, pp.370~383,1980.

4. Huan, J. S. and T. L. James, and P. W. Trefor, “Using Neural Networks to Predict Component Inspection Requirements for Aging Aircraft,” Computers Ind. Engng, Vol .30, No.2, 257-267,1996.

5. Holt,CC., “Forecasting Seasonal and Trends by Exponentially Weighted Moving Averages,” Office of Naval Research, Memorandum,52, 1957.