- 學位論文
應用移動拔靴法與倒傳遞網路於備用零件最適需求預測模式之研究
Applying Moving Bootstrap and Back-Propagation Neural Network for the Optimization Demand Forecasting Model of Spare Parts
摘要
對大多數企業而言，存貨是最昂貴的資產之一，特別是備用零件的部分，而備用零件的需求一般來說屬於間歇性需求，過往的研究指出拔靴法在處理間歇性需求的問題上有較佳的預測效果，但備用零件仍有部分屬於規律的需求，在該需求型態上，使用移動平均法普遍會有不錯的預測效果。本研究以拔靴法為基礎提出移動拔靴法，並配合移動平均法建立組合預測模式，依據各品項之預測誤差，找出較佳的預測模式。最後再使用倒傳遞網路建立可自動選取預測模型的分類模式，透過該模式本研究找出了「日平均耗用量」、「耗用量為零之天數比例」及「日耗用量標準差」三項可用以需求預測模式配適度分類的關鍵變數，未來企業欲增添新備用零件時，可以依照該分類模型進行判斷，將有助於降低預測的誤差，達到庫存成本降低與營運績效提升之目標。
並列摘要
Inventory control of spare parts has been an essential to many organizations since it is one of the most expensive assets. Most of the spare parts are to belong to intermittent demand and Bootstrapping has been claimed to be of great value for forecasting. While a small proportion of spare parts are regard to regular demand, Moving Average, frequently used to deal with this type of demand. We address combination forecasting model by the Moving Bootstrap based on Bootstrapping and Moving Average to classify the appropriate method. Then use the Back-Propagation Neural Network to construct the classification model which can be used to automatically select the better approach of forecasting. We find that the main explanatory variables about consumption of daily average, the ratio of days with zero consumption and standard deviation of daily consumption can exact classify the demand forecasting approach. In the future, enterprise arranges to purchase new spare parts, this combination model will assist in concluding the forecasting method and reducing the forecast error. Moreover, it leads to lower stock costs and improves operational performance.