隨著臺灣「強韌電網建設計畫」推動，智慧電錶需求快速成長，然而全球半導體短缺與供應鏈不確定性，使電子業關鍵零組件面臨高度缺料風險。傳統物料需求規劃（Material Requirements Planning， MRP）建立於需求穩定假設，難以因應動態波動，使企業長期處於被動補貨困境。為突破此一侷限，本研究以智慧電錶製造為個案，結合MRP資料與深度學習建構缺料預測模型。研究以再訂購點差距（Reorder Point Gap， ROP Gap）量化缺料風險邊界，運用深度神經網路（Deep Neural Network， DNN）建立紅黃綠（Red-Yellow-Green， RYG）三色預警機制，並比較均方誤差（Mean Squared Error， MSE）與交叉熵（Cross Entropy）兩種損失函數之表現。實驗顯示，Cross Entropy模型在F1-score與風險類別（R 類）召回率上皆優於 MSE 模型，對關鍵風險類別達成零漏判（R、Y 類誤判為安全類別 G 之筆數皆為0），並為D公司節省誤判成本比例達80%。另外，透過SHAP（SHapley Additive exPlanations）分析，識別出再訂購點差距、庫存為缺料風險的關鍵因子。本研究亦驗證MRP與DNN之互補性，兩者為「規劃-預測」分工架構，使企業由被動補貨轉向主動預防，有效降低缺料所致營運風險與成本，提升智慧電錶製造之供應鏈韌性。;With the implementation of Taiwan′s "Resilient Power Grid Construction Plan，" demand for smart meters has grown rapidly. However， the global semiconductor shortage and supply chain uncertainties have exposed critical components in the electronics industry to a high risk of material shortages. Traditional Material Requirements Planning （MRP） is built on the assumption of stable demand and struggles to respond to dynamic fluctuations， leaving enterprises trapped in a long-term cycle of reactive replenishment. To overcome this limitation， this study takes smart meter manufacturing as a case and integrates MRP data with deep learning to construct a material shortage prediction model. The Reorder Point Gap （ROP Gap） is used to quantify the boundary of shortage risk， and a Deep Neural Network （DNN） is applied to establish a Red-Yellow-Green （RYG） three-tier early-warning mechanism. The performance of two loss functions， Mean Squared Error （MSE）and Cross Entropy， is further compared. The experimental results show that the Cross Entropy model outperforms the MSE model in both F1-score and recall for the risk class （R class）， achieving zero missed detection for critical risk classes （no R or Y class samples were misclassified as the safe class G）， and reducing D Company′s misclassification cost by 80%. Furthermore， SHAP （SHapley Additive exPlanations） analysis identifies the Reorder Point Gap and stock level as the key factors influencing material shortage risk. This study also verifies the complementarity between MRP and DNN， forming a "planning–prediction" division of labor that enables enterprises to shift from reactive replenishment to proactive prevention， effectively reducing operational risks and costs caused by material shortages and enhancing the supply chain resilience of smart meter manufacturing.
