喪失銷貨存貨系統經驗訂購政策之研究

Abstract

大多數存貨系統假設需求若無法立即滿足可後補。實務上需求若無法立即滿足常喪失。與後補存貨模式之研究相比，喪失銷貨存貨模式之文獻較少。本計畫探討前置時間為正數但短於盤存周期之喪失銷貨存貨系統。2006 年姜齊已針對此存貨系統研擬出最適訂購政策，但使用需回復計算的動態規劃方法。如Silver 於2008 年指出，存貨管理理論與實務存在明顯差異，實務上很少人使用難於執行的方法，反之他們使用經驗或簡而易懂的政策。本計畫針對上述存貨系統提出兩個經驗訂購政策: 一為訂購後水準政策，另一為修正固定訂單政策。訂購後水準政策為後補存貨模式之最佳政策，而固定訂單政策近來為一些學者所研究。本計畫研擬一個設定訂購後水準的方法，並比較上述兩個政策與最佳訂購政策的績效。研究結果顯示所提出的兩個經驗政策為不錯之訂購政策，特別是修正固定訂單政策在大多數情況下表現優於訂購後水準政策。實務界人士因此至少有一個簡單易執行的訂購政策可使用。

Most inventory systems assume that demand not filled immediately is backlogged. In practice, demand not filled at once is often lost. The literature on lost-sales inventory models is scarce, compared to the studies on backlogged inventory models. In this research, we consider lost-sales periodic review inventory systems where the lead-time is positive but shorter than the period length (for example, the lead-time is two or three days while the review period is one week). Chiang [European Journal of Operational Research, 170 (2006) 44-56] actually has developed optimal ordering policies for such systems. However, he used the dynamic programming approach which entails recursive computation. As noted by Silver [Infor, 46 (2008) 15-27], there is a significant gap between theory and practice in inventory management. Practitioners usually do not use difficult-to-program (though optimal) methods； instead, they prefer heuristic (though not optimal) ordering policies. In this research, we propose two heuristic ordering policies for the lost-sales periodic review systems: one is an order-up-to policy and the other is a revised constant-order policy. The order-up-to policy is known to be optimal for the backlogged inventory models, while the constant-order policy was recently examined by various scholars. In this research, we devise a method to set the order-up-to level for the lost-sales models. We also compare the performance of these two heuristics with the optimal ordering policy proposed by Chiang. We show that these two heuristics perform well under a wide range of input parameters. In particular, the revised constant-order policy performs better than the order-up-to policy under most parameter settings. Thus, practitioners has at least a simple ordering policy to use.