備用零件之最後訂購數量模型

Abstract

隨著生活水準提升，對於產品的要求也逐漸提高，生產的品質改善和價格的折扣已無法滿足消費者，使得廠商不僅必須在生產產品的品質、物流的速度上競爭，也逐漸將重點轉移到售後服務上。由於汽車銷售的時間往往比維修服務的時間來的短，因此在汽車停產之後，維修零件往往必須繼續提供，但是此時由於汽車已經停產，許多零件的型號會逐漸被淘汰，而零件需求量也會逐漸降低，因此，零件供應商在考慮成本因素下，將不會繼續生產零件到服務終止，此時代理商面臨上游供應商即將停產某些維修零件的情況，必須提出最後一次訂購以滿足零件在衰退期的剩餘需求。由於代理商必須預防存貨過剩或存貨不足的情形，所以針對最後訂購數量的決策必須慎重決定。 因此，本研究希望能夠提出一個較佳的零件需求預測模型，以協助代理商面對最後訂購數量的決策。本研究提出兩種預測模型，分別為 Regression 模型和 Installed Base 模型，以 all-time requirement 的概念，透過從最後訂購時間點到服務終止時間點的未來零件需求預測，進行備用零件的存貨，並透過誤差百分比進行衡量模型，便可以將這兩種模型與個案公司的計算法進行比較。最後，透過改變最後訂購時間點，比較預測模型和個案公司計算法在不同最後訂購情境下的預測表現，期望找出一套最有效的零件需求預測模式。如此，便可以提供個案公司在面對不同的最後訂購情境下，應如何選擇較佳的模型才能做出更具成本效益的訂購決策，以有效滿足顧客在剩餘服務時間的零件需求。

With the living standard elevating, it is hard to satisfy consumers with simply improvements of production quality and price discounts. The manufacturers not only focus on the production quality and the speed of logistics, but also put more emphasis on after-sales services. Since the periods of maintenance and replacements of spare parts are much longer than the vehicle’s production periods, there is need to stock up the parts for a certain period of time even after the vehicle manufacturer stops producing the car. However, as the car sales discontinue, the suppliers will stop providing parts due to the reducing demands. As a result, the agent will face the difficulties on making precise decisions on the quantity of final orders, which are expected to last until all service contracts have ended. The agent should make a precise final order quantity decision to meet part demands at the end of their lifetimes. In this paper, we want to find a better forecasting model, to help the agent find the optimal final order quantities. There are two forecasting models to deal with final order problems - the Regression model and the Installed Base model. With the concept of all-time requirements, the real data from company H are used to make a stock prediction of spare parts from the last order point to the end point of services. By comparing the percentage of forecast error, it is known which model performs better. Finally, by changing the decision point of final order, we compare the performance of our forecasting model with the one from company H. With this robustness test, we can help company H to find the best forecasting model under different final order conditions.