短期訂單預測模型之研究—PDA產業為例

A Study on the Short-Term Sales Forecasting-An Example of PDA Industry

羅慕君(MU-CHUN LO)

指導教授 ： 許通安

中原大學/商學院/資訊管理研究所/碩士(2004年)

https://doi.org/10.6840/cycu200400602

摘要

近年來消費型科技的日新月異，PDA由當初單純的記事功能，逐漸演變至多功能，含手機功能的PDA，將如同手機般成為消費性產品，產品的生命週期縮短，小量多樣的產品勢必成產業的特性。如何有效的預測訂單，使得企業即時反應市場動態，將成為企業成功的關鍵性指標。
本研究僅就訂單預測（Sales Forecasting）中，有關短期訂單之部份做探討，因訂單預測為所有商業活動預測之源頭，且短期訂單佔45%營業額，又不如長期訂單有合約做規則性計劃。針對PDA產業的了解，分析訂單預測在產業中的管理地位。
本研究將以PDA產業為對象，討論移動平均法、指數平滑法、倒傳遞類神經網路、灰色預測四種預測方法，運用在不同性質之產品下所產生之預測效果，及其適用性。針對短期之訂單預測所可能遇到 的問題逐一表述，並將此四種預測方法分別進行實驗與討論。
實驗發現，樣本資料如易受市場流行驅勢波動、易受季節性影響之產品，不適合使用灰預測、移動平均、指數平滑法…等，倒傳遞類神經網路在各類型資料的預測上效果較佳。故建議使用倒傳遞類神經網路來建立預測模型。

關鍵字︰預測模型、移動平均法、指數平滑法、倒傳遞類神經網路、短期訂單、Matlab。

關鍵字

短期訂單 ； 預測模型 ； 移動平均法 ； Matlab ； 指數平滑法 ； 倒傳遞類神經網路

並列摘要

For the latest years, the technology of consumer product has been ever newer, and PDA, as one of them, has been continuously improved from easy-to-use simple function—personal data assistant— to the one with solutions for manipulating the full range of multimedia resources. PDA, with the convergence of data-centric and voice-centric functions, is becoming one of the consumer products gradually as the cellular phones nowadays. Besides, as such product life cycle has been ever shorten, the product with small quantity but various choices will definitely lead the future market in this industry. How to have the sales forecast effectively to react the market trends just in time will be the key target for the enterprise to success.
This research focuses on short-term orders in Sales Forecasting. Sales Forecast is the beginning of the prediction in all business activities, especially for the short-term orders which obtains 45% in business volume that unable to set a regular planning compared to long-term ones which has the contract to follow. Therefore, based on the understanding of PDA industry, this study analyzes the sales forecasting in the place of management in the industry.
Aiming at PDA industry, this research discusses specifically four ways to set the forecasting, the Moving Average Models, the Exponential Smoothing Method , the Back Propagation Neural Networks method, and the Grey method that applies on different products to see the effects and the results, as long as to explore all the potential questions in dealing with the sort-term forecasting during the application process with four methods mentioned above individually. Through these experimentals, as a result, the sample data shows that if the provided data is easily flattered by the market trends, or the product itself is the seasonal product may not be applied to either Grey method, the Moving Average Models, or the Exponential Smoothing Method to process the forecasting. However, taking the method of Back Propagation Neural Networks seems to have a better effect result in forecasting. As a consequent, this study will suggest to use the method of Back Propagation Neural Networks to set up the model of forecasting.
Key words: Forecasting model, Moving Average method, Exponential Smooth- ing method, Back Propagation Neural networks method, Short-term order,Matlab.

並列關鍵字

Back Propagation Neural networks method ； Forecasting model ； Moving Average method ； Matlab ； Short-term order ； Exponential Smooth- ing method

參考文獻

[ 4 ] Kimoto, T., and Asakwa, K. (1990), "Stock market predication system with modular networks" , IJCNN-90-Wash., I, 1-6.

[ 5 ] Ramanujam, J., and Sadayappan, P. (1988), "Optimization by neural network", ICNN-88, II, 325~332. 註

[ 6 ] Sharda, R. and Patil, R. B. ，1990，"Neural networks for forecasting experts：an empirical test", IJCNN-90-Wash., II, 491-494.

[ 27 ] 周湘蘭 (2001)，類神經網路在多重產品需求預測上之應用，元智大學，工業工程管理研究所，碩士論文。

[ 1 ] Baba, K. and Kozaki, M. (1992), "An intelligent forecastion system of stock price using neural network" , IJCNN-92, I, 371-377. 註