倒傳遞網路在產品需求預測之研究─以模組化產品為例

The Implementation of Back-Propagation Network in Demand Forecasting Model ─ A Case for Modular Products

鄭勻惠(Yun-Hui Cheng)

指導教授 ： 陳雲岫

元智大學/工程學院/工業工程與管理學系/碩士(2004年)

摘要

近年來，企業經營環境的變化與消費者需求型態改變，使得企業為了增加其市場競爭優勢，必須提高顧客滿意度及減少製造成本。因此決策者必須規劃適當生產計劃和安全存貨量，才能降低缺貨成本、延遲成本及存貨成本，然而預測的準確與否往往會影響製造成本及決策的品質。當決策者在做短期生產計畫時，可能會因短期需求波動及產品具有相關性，而影響需求預測的準確性。因此，考量模組化產品間關連性，建構出適合生產線需求預測模式為一重要研究課題。
本研究運用倒傳遞網路在預測上優越能力，學習模組化產品關連性進行需求預測模式之建構，以供決策者執行生產規劃之依據。研究中，利用AweSim模擬軟體，分別模擬六種及九種零組件之六個月短期訂單，收集訂單到達時間、種類及數量，以BPN網路進行需求預測模型之建立。需求預測模式利用移動視窗法進行網圖建構，其輸入及輸出變數為訂單到達之間隔時間、產品數量、訂單種類權重，以平均方根誤差(MSE)作為評估指標。研究結果顯示，運用BPN網路建構六種及九種零組件模組化產品之需求預測模式，在訂單分配率改變時，BPN網路仍有不錯的預測能力，其最佳需求預測模式之測試樣本平均MSE值皆在0.2以下。

關鍵字

生產計劃 ； 模組化產品 ； 需求預測 ； BPN網路

並列摘要

In recent years, the business operating environment varies and consumer demanding follows the trend of globalization, the business must raise customer satisfaction and reduce the manufacturing cost in order to increase its market competitive ability. So the decision maker must process an adequacy production planning and reach safety stock, meanwhile reduce the shortage cost, delay cost and stock cost. However the accuracy of forecast usually influences the manufacturing cost and decision quality. When the decision maker is doing short-term production planning, the short-term demand fluctuation and product relativity may influence the demand forecast accuracy. Therefore, scheduling the modular product to fit the production line shot-term demand-forecasting model is important and necessary.
In this thesis, we take the merit of the Back-Propagation neural network (BPN) in superiority ability of the forecasting to the modular products system and establish a demand-forecasting model, this model can provide the decision maker carries out the basis of production planning. In this research, we use the simulation software AweSim to simulate the orders data in six kinds and nine kinds of components for six months. Data for order types, arrival times, and numbers of product are collected and trained by the BPN to build a BPN modular product demand-forecasting model. The BPN demand-forecasting model is constructed by the moving windows method and average square error (MSE) is the evaluation index for demand-forecasting model. Simulation study shows that the BPN forecasting model performance good on the different order distributions, the average mean square errors (MSE) of the best model test samples are all under 0.2.

並列關鍵字

Production Planning ； Modular Products ； Demand Forecasting Model ； Back-Propagation Neural Network

參考文獻

1. Alvarez, J. A., D. Carmona, M. A. Jaramillo and E. Gonzalez, “Electric energy demand forecasting with neural networks,” IEEE 28th annual conference of the Industrial electronics society, Vol.3, pp.1860-1865, Nov. 2000.

2. Charytoniuk, W. and M. S. Chen, ” Neural-network-based demand forecasting in a deregulated environment,” IEEE Transactions on industry applications, Vol. 36, No. 3, pp.893-898, May/June 2000.

3. Desouky, A. A. EI and M. M. EIKateb, “ Hybrid adaptive techniques for electric-load forcast using ANN and ARIMA,” IEE Proc-Gener. Transm. Distrib, Vol. 147, NO. 4, pp.213-217, July 2000.

4. Dantas, A., K. Yamamoto, M. V. Lamar, and Y. Yamashita, “Neural network for travel demand forecasting using GIS and Remote Sensing,” IEEE proceeding, pp.435-440, 2000.

5. Gardner Jr., E. S. and D. S. Joaquin, “Seasonal adjustment of inventory demand series: a case study,” International Journal of forecasting, Vol. 18, pp. 117-123, 2002