面對全球永續與碳中和目標，企業急需於產品設計階段導入低碳策略。桌上型電腦作為高碳電子產品，其零組件在材料與製造等環節皆會影響碳排放，而現行碳排放大多於生產後評估，無法於設計初期即時預測。 本研究提出一套深度學習預測模型，結合自編碼器進行特徵選取、類神經網路進行碳排放回歸預測，並導入知識蒸餾以強化模型精準度與效能。研究建立桌上型電腦零組件碳排放資料集，並於高效能、商務型與輕量型產品中進行模型驗證。 實驗結果顯示，該模型可於設計初期有效預測各組件碳排放，協助企業優化設計並落實低碳設計目標，對電子產業永續發展具有高度實務應用價值。;In the face of global sustainability and carbon neutrality goals, enterprises urgently need to implement low-carbon strategies during the product design stage. Desktop computers, as high-carbon electronic products, generate emissions throughout material selection and manufacturing processes. However, current carbon footprint assessments are mostly conducted post-production, lacking the ability to predict emissions early in the design phase. This study proposes a deep learning-based predictive model that integrates autoencoders for feature selection, neural networks for carbon emission regression, and knowledge distillation to enhance model accuracy and performance. A carbon emission dataset of desktop computer components was constructed and the model was validated across high-performance, business, and lightweight product categories. Experimental results show that the proposed model can effectively predict component-level carbon emissions during the early design phase, assisting enterprises in optimizing product design and achieving low-carbon goals. This approach provides strong practical value for promoting sustainability in the electronics industry.
