摘 要 當企業必頇對擁有較長生命週期的非消耗性產品提供維修售後服務時,對企業而言,其零件的存貨管理是一個沈重的負擔。因為維修零件隨機性出現的需求與其不穩定的數量,令零件需求預測難以進行。尤其是當發生零件停止供應時,企業需要預測未來數年的零件需求數量,以應對之後可能發生的維修需求。此一最終訂購問題,目前較少有相關文獻對於需求預測方面加以探討,因此,本研究乃針對零件需求的長期預測建立數學模式並加以求解。 本研究期望可建構令「最終訂購問題」的預測發生誤差為最小的模式,並使用最小平方法建立目標函數,由於本研究所提出之問題是屬於非凸集合問題,因此本研究採取不易落入局部最佳解之基因演算法求解,並利用A公司零件歷史資料庫進行實證研究與分析。 為了測詴本研究建構之預測模式的求解成效,本研究另外採取一般常用的「移動平均法」與「指數平滑法」做為比較之基礎,將其求解結果與實際需求量進行比較。其結果顯示,預測模式的預測誤差皆少於採用時間序列法的預測誤差0.2~135%左右。且在面對零件需求有著較大的波動起伏時,其預測能力較時間序列法有更好的表現。
ABSTRACT It is a burden in inventory management for enterprises when they need to provide maintenance and after-sale services for non-consumable products that have longer life cycles. The random demand and uncertainty for spare parts make future prediction hardly proceed, particularly when the parts production stop. Therefore, enterprises must make proper prediction for the parts meet future needs. “Final-order” issue in prediction demand rarely discussed in past. The purpose of this study tries to solve this issue by constructing a mathematical model for long-term prediction of spare parts. This study is expected to find out a model that has minimum error in predicting the demand of spare parts. Due to the study is not able to be proved as convex set, Genetic Algorithm, it is not easy to fall in local optimum, is applied to this study. This study applied the data of company A to prediction and numerical analysis. This study applies “moving average method” and “exponential smoothing method” to compare with this model. The results show the errors in prediction of this model are better than time series methods which errors range are between 0.2 to 150%. Also, this model shows better performance while there is a great variation demand.