請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88588
標題: | 以時序圖神經網路進行半導體產業供應鏈短缺預測 Supply Chain Shortage Forecasting for Semiconductor Industry by Temporal Graph Neural Network |
作者: | 簡辰安 Chen-An Chien |
指導教授: | 陳建錦 Chien Chin Chen |
關鍵字: | 時序圖神經網路,供應鏈短缺,半導體產業,存貨天數,深度學習, temporal graph neural network,supply chain shortage,semiconductor industry,days of inventory,deep learning, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 半導體產業的高成本、垂直分工的供應鏈及短暫的產品生命週期使其非常仰賴穩定的供應鏈,供應鏈的延遲可能會導致生產停滯與伴隨而來的財務損失。為了使半導體公司得以即使依照上游供應情況調整生產計劃,準確預測供應是否有短缺的風險至關重要。本論文提出了一種供應鏈短缺預測模型,為了透過時序圖神經網路 (TGNN) 分析財務與文字資料,我們結合了存貨週期 (DOI) 和法說會逐字稿,並以實際案例證明其可以用於供應短缺預測。我們將供應鏈構建成一個網路,藉由考量上下游供應商之間的關係,試圖捕捉當上游供貨不穩定時,下游被波及的可能性與時間。為了驗證模型,我們收集了2018年至2022年的五年間台灣積體電路製造股份有限公司 (TSMC) 的數據,其中包含供應商的供應鏈關係、財務數字及法說會逐字稿,並以實驗證明了機器學習模型在預測供應短缺方面的有效性,還有法說會逐字稿可以做為輔助潛在短缺預測的指標。本研究開啟了將財務數字、文字及供應鏈關係整合到TGNN中以預測供應短缺的可能性,為半導體供應商供應短缺提供了一種解決方法,並具有未來應用和研究的潛力。 The semiconductor industry heavily relies on a well-functioning and efficient supply chain due to its high cost, vertically divided supply chain, and short product life cycles. Delays in the supply chain can lead to production stagnation and associated financial losses. Accurately forecasting supply shortages is crucial for semiconductor companies to adjust their production plans based on upstream supply conditions. This master thesis proposes a supply chain shortage forecasting model that utilizes a Temporal Graph Neural Network (TGNN) to analyze financial and textual data. We incorporate Days of Inventory (DOI) and earnings call transcripts to predict potential supply shortages. By constructing a supply chain network that considers the relationships between upstream and downstream suppliers, we aim to capture the likelihood and timing of downstream disruptions when upstream supply is unstable. To validate the model, we collect the experimental data from Taiwan Semiconductor Manufacturing Company (TSMC) spanning a five-year period from 2018 to 2022, including suppliers' supply chain relations, financial numbers and earnings call transcripts. The experiments demonstrate the effectiveness of machine learning models in forecasting supply shortages, and earnings call transcripts serve as supplementary indicators for potential shortages. As a result, this research is the first work to integrate financial figures, textual data, and supply chain relationships into TGNN, providing a solution for predicting supplier shortages in the semiconductor industry and showing potential for future applications and research. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88588 |
DOI: | 10.6342/NTU202301990 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 資訊管理學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf | 2.8 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。