生產模式與基本庫存策略之研究：考量不確定性因子

Abstract

近年來，供給端和需求端的不確定性是企業運作的兩大難題，能夠以較低成本、較快速度滿足顧客需求的組織，就更有競爭優勢。一項產品可以有不同的BOM表設計，其對應的生產模式也會有所不同，因此本研究比較不同的BOM表設計和生產模式，在需求、訂購頻率、前置時間的不確定性下，制訂出對整體供應鏈最佳的基本庫存策略。 本研究提出一個隨機模型，並以模擬的方法比較不同生產模式和存貨政策所造成的影響。然而，隨著問題規模趨於龐大，所需的搜尋範圍將大幅增加，若以全域搜尋的方式尋找解答，需耗費大量的時間以及計算資源，因此，本研究提出一個啟發式演算法，使得本研究問題在有效率的時間下，得到趨近最佳解之可行解決方案。 本研究啟發式演算法流程可分為三步驟：定義領導者、領導者和追隨者的基本庫存制訂、模擬以及搜尋最佳解。第一步，根據供應鏈成員的特性決定出一個領導者，而其他的成員則定義為追隨者。第二步，利用領導者的成本結構制訂出基本庫存策略的搜尋範圍，而追隨者採用和領導者一樣的機制。第三步，根據所訂出基本庫存策略的搜尋範圍進行模擬流程，並搜尋最佳解。雖然已事先訂定出基本庫存策略的搜尋範圍，但要每一個組合都進行模擬流程將會耗費大量時間以及資源，因此本研究不採用全域搜尋的方式，而使用三層式搜尋法有效率地解決本研究問題。最後，本研究實作出此模擬系統，並進行情境分析，比較不同因子的設定對生產模式規劃的影響。本研究利用實際案例測試，驗證本演算法確實可行且具高效率性。

Recent years, variations and uncertainties of supplies and demands are two main difficulties in doing businesses for many companies. An organization who satisfies customer demand in lower cost and faster speed gains the upper hand. A product can have different designs of BOM, which in turn correspond to different kinds of production environments. In this study, we compare the different BOM designs and production environments, and determine the optimal base-stock policy for each supply chain member who is allowed to stock inventories and compare the profits generated by MTS and ATO under the uncertain conditions such as demand, demand frequency, and lead time. This study formulates a stochastic model and solves the model using the simulation method to compare the impacts caused by the different production environments and different inventory policies. However, as the problem size increases, the search range grows exponentially. It becomes impractical to conduct a global search due to the considerable time and computer resources. Therefore, this study proposes a heuristic algorithm, called the Leader’s Base-Stock Policy Algorithm (LBSPA) to solve this problem effectively. The main process of the algorithm in this study can be divided into three phases: Leader Finding, Leader and Followers’ (R, Q) setting, Simulation and searching a proper solution. In Leader Finding, we classify the supply chain members into leaders and followers, and develop ruled based selecting mechanisms to identify the leader. The rest of the supply chain members are defined as followers. In Leader and Followers (R, Q) setting, we define the search range of the leader’s base stock policy and followers’ (R, Q) which are based on leader’s policy. In the last step, we run the simulation process according to the base stock policies which are defined in advance, and search the proper solution. Although we have defined the ranges of R and Q, it takes lots of time to simulate every single combination of R and Q. Instead of conducting a global search method, this study develops three-level interval search to solve the problem more efficiently. To show the effectiveness and efficiency of LBSPA, a prototype is constructed and a scenario analysis is conducted.