Publications-Theses
Article View/Open
Publication Export
-
題名 機器學習於高變異、不平衡的生產需求預測
Machine Learning for Demand Forecasting with Highly Volatile and Imbalanced Data作者 蔡睿峰
Tsai, Jui-Feng貢獻者 莊皓鈞<br>周彥君
Chuang, Hao-Chun<br>Chou, Yen-Chun
蔡睿峰
Tsai, Jui-Feng關鍵詞 電子零組件供應鏈
需求預測
機器學習
Semiconductor Supply chain
Demand Forecast
Machine Learning日期 2020 上傳時間 2-Sep-2020 11:45:11 (UTC+8) 摘要 電子產品蓬勃發展連同帶動半導體產業發展，相關的電子零組件供應變成為該產業鏈的重要一環。零組件通路商更是高科技半導體供應鏈的協調角色，如何控制需求與供給是零組件通路商在經營管理上的一大課題。因此，導入需求預測可以幫助通路商降低存貨成本，且兼顧服務水準。一般來說，半導體通路商的需求預測會有以下問題，半導體零組件的種類繁多，零組件產品生命週期不固定，前置時間長等。半導體零組件常見的需求分布有以下三種，「產品生命週期長，需求不固定，有尖峰情形」，「產品需求頻繁，但高度變異」以及「產品生命週期長短不一」。因此本研究期望透過機器學習(Machine Learning)的技術，解決半導體供應鏈上跨品項廠區的需求預測問題。本次研究與亞洲一知名的半導體零組件通路商合作，使用該公司內部系統2017年至2018年共兩年的原始資料作為研究基礎，根據預設情境設定前置時間為12周的需求預測，就原始高變異、不平衡的資料進行Temporal Aggregation，以及跨品項廠區的資料前處理與資料特徵工程，並使用Random Forest及XGBoost等集成式機器學習(Ensemble Learning)模型，配合參數調整來分析單一機器學習模型的預測效果。此外，針對時間序列內12周需求為0的部分，設計一個二階段的預測方式，進而提升模型的預測效果。在探索性研究中，加入分群方法(Clustering)，使用零組件產品過往的拉貨資訊將資料做分群，區分出拉貨量異常的資料樣本。本研究發現，在預設情境前置時間12周的需求預測。搭配二階段模型可以提升單一模型的預測效果，探索性研究的分群方法可以區分拉貨量異常的資料樣本。提供半導體零組件通路商一個需求預測的參考方式。
A rapid growth of the consumer electronics market has led the semiconductor industry flourished. The supply of related components has become an important part of this industrial chain. Especially, the distributor plays a key role in this supply chain because it has to match demand from downstream manufacturers with supply from upstream vendors. How to forecast uncertain demand has become a critical task for a distributor’s operations. Related studies suggest that it is common for a semiconductor distributor to face demand uncertainty from lots of components with non-homogeneous product lifecycle and long supply lead time. Under volatile, sporadic, and unbalanced demand patterns, elevating demand forecast accuracy is crucial for a distributor to maintain high service level while lowering inventory holding cost. This study investigates the use of contemporary machine learning methods to help the distributor tackle challenging demand forecasting problems.We work with a leading electronics distributor in Asia and perform analysis using a large dataset over 2017 to 2018 from the company. The goal is to improve prediction accuracy of total demand over a 12-week lead time for many items. We first employ temporal aggregation to filter non-smooth demand and derive more a wide array for predictor variables through feature engineering. We then employ two state-of-the-art ensemble learning algorithms – Random Forest and XGBoost – to predict demand. Inspired by existing studies on intermittent/erratic demand modeling, we propose a two-stage model grounded on XGBoost. We show this model greatly improves overall prediction performance over ordinary ensemble learning. We further conduct an exploratory research where we apply k-means clustering to identify outlying demand observations.參考文獻 Breiman, L. (2001). Random forests. Machine Learning, vol.45, 5–32.Carbonneau, R., Laframboise, K., & Vahidov, R. (2008). Application of machine learning techniques for supply chain demand forecasting. European Journal of Operational Research, vol.184(3), 1140–1154.Chen, T., & Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System.Friedman, J. (2001). Greedy Function Approximation : A Gradient Boosting Machine. The Annals of Statistics, vol.29(5), 1189–1232.Hartigan, J. A., & Wong, M. A. (1979). A K-Means Clustering Algorithm. Journal of the Royal Statistical Society. Series C (Applied Statistics), vol.28(1), 100–108.Ma, S., & Fildes, R. (2020). Forecasting third-party mobile payments with implications for customer flow prediction. International Journal of Forecasting.Nikolopoulos, K., Syntetos, A. A., Boylan, J. E., Petropoulos, F., & Assimakopoulos, V. (2011). An aggregate-disaggregate intermittent demand approach (ADIDA) to forecasting: An empirical proposition and analysis. Journal of the OperationalQu, T., Zhang, J. H., Chan, F. T. S., Srivastava, R. S., Tiwari, M. K., & Park, W. Y. (2017). Demand prediction and price optimization for semi-luxury supermarket segment. Computers and Industrial Engineering, vol.113, 91–102.Rostami-Tabar, B., Babai, M. Z., Syntetos, A., & Ducq, Y. (2013). Demand Forecasting by Temporal Aggregation. Naval Research Logistics, vol.60(6), 479–Syntetos, A. A., Babai, Z., Boylan, J. E., Kolassa, S., & Nikolopoulos, K. (2016). Supply chain forecasting: Theory, practice, their gap and the future. European Journal of Operational Research, vol.252(1), 1–26.Syntetos, A. A., & Boylan, J. E. (2005). The accuracy of intermittent demand estimates. International Journal of Forecasting, vol.21(2), 303–314.Teunter, R. H., Syntetos, A. A., & Babai, M. Z. (2011). Intermittent demand: Linking forecasting to inventory obsolescence. European Journal of Operational Research, vol.214(3), 606–615.Vairagade, N., Logofatu, D., Muharemi, F., & Leon, F. (2019). Demand Forecasting Using Random Forest and Artificial Neural Network for Supply Chain Management. International Conference on Computational Collective Intelligence, 328–339.Willemain, T. R., Smart, C. N., Shockor, J. H., & DeSautels, P. A. (1994). Forecasting intermittent demand in manufacturing: a comparative evaluation of Croston’s method. International Journal of Forecasting, vol.10(4), 529–538. 描述 碩士
國立政治大學
資訊管理學系
107356006資料來源 http://thesis.lib.nccu.edu.tw/record/#G0107356006 資料類型 thesis dc.contributor.advisor 莊皓鈞<br>周彥君 zh_TW dc.contributor.advisor Chuang, Hao-Chun<br>Chou, Yen-Chun en_US dc.contributor.author (Authors) 蔡睿峰 zh_TW dc.contributor.author (Authors) Tsai, Jui-Feng en_US dc.creator (作者) 蔡睿峰 zh_TW dc.creator (作者) Tsai, Jui-Feng en_US dc.date (日期) 2020 en_US dc.date.accessioned 2-Sep-2020 11:45:11 (UTC+8) - dc.date.available 2-Sep-2020 11:45:11 (UTC+8) - dc.date.issued (上傳時間) 2-Sep-2020 11:45:11 (UTC+8) - dc.identifier (Other Identifiers) G0107356006 en_US dc.identifier.uri (URI) http://nccur.lib.nccu.edu.tw/handle/140.119/131488 - dc.description (描述) 碩士 zh_TW dc.description (描述) 國立政治大學 zh_TW dc.description (描述) 資訊管理學系 zh_TW dc.description (描述) 107356006 zh_TW dc.description.abstract (摘要) 電子產品蓬勃發展連同帶動半導體產業發展，相關的電子零組件供應變成為該產業鏈的重要一環。零組件通路商更是高科技半導體供應鏈的協調角色，如何控制需求與供給是零組件通路商在經營管理上的一大課題。因此，導入需求預測可以幫助通路商降低存貨成本，且兼顧服務水準。一般來說，半導體通路商的需求預測會有以下問題，半導體零組件的種類繁多，零組件產品生命週期不固定，前置時間長等。半導體零組件常見的需求分布有以下三種，「產品生命週期長，需求不固定，有尖峰情形」，「產品需求頻繁，但高度變異」以及「產品生命週期長短不一」。因此本研究期望透過機器學習(Machine Learning)的技術，解決半導體供應鏈上跨品項廠區的需求預測問題。本次研究與亞洲一知名的半導體零組件通路商合作，使用該公司內部系統2017年至2018年共兩年的原始資料作為研究基礎，根據預設情境設定前置時間為12周的需求預測，就原始高變異、不平衡的資料進行Temporal Aggregation，以及跨品項廠區的資料前處理與資料特徵工程，並使用Random Forest及XGBoost等集成式機器學習(Ensemble Learning)模型，配合參數調整來分析單一機器學習模型的預測效果。此外，針對時間序列內12周需求為0的部分，設計一個二階段的預測方式，進而提升模型的預測效果。在探索性研究中，加入分群方法(Clustering)，使用零組件產品過往的拉貨資訊將資料做分群，區分出拉貨量異常的資料樣本。本研究發現，在預設情境前置時間12周的需求預測。搭配二階段模型可以提升單一模型的預測效果，探索性研究的分群方法可以區分拉貨量異常的資料樣本。提供半導體零組件通路商一個需求預測的參考方式。 zh_TW dc.description.abstract (摘要) A rapid growth of the consumer electronics market has led the semiconductor industry flourished. The supply of related components has become an important part of this industrial chain. Especially, the distributor plays a key role in this supply chain because it has to match demand from downstream manufacturers with supply from upstream vendors. How to forecast uncertain demand has become a critical task for a distributor’s operations. Related studies suggest that it is common for a semiconductor distributor to face demand uncertainty from lots of components with non-homogeneous product lifecycle and long supply lead time. Under volatile, sporadic, and unbalanced demand patterns, elevating demand forecast accuracy is crucial for a distributor to maintain high service level while lowering inventory holding cost. This study investigates the use of contemporary machine learning methods to help the distributor tackle challenging demand forecasting problems.We work with a leading electronics distributor in Asia and perform analysis using a large dataset over 2017 to 2018 from the company. The goal is to improve prediction accuracy of total demand over a 12-week lead time for many items. We first employ temporal aggregation to filter non-smooth demand and derive more a wide array for predictor variables through feature engineering. We then employ two state-of-the-art ensemble learning algorithms – Random Forest and XGBoost – to predict demand. Inspired by existing studies on intermittent/erratic demand modeling, we propose a two-stage model grounded on XGBoost. We show this model greatly improves overall prediction performance over ordinary ensemble learning. We further conduct an exploratory research where we apply k-means clustering to identify outlying demand observations. en_US dc.description.tableofcontents 第一章 緒論 1第一節 研究背景與研究動機 1第二節 研究問題 3第二章 文獻回顧 5第三章 資料與方法 8第一節 資料與情境敘述 8第二節 模型與方法 16第四章 預測結果 22第五章 探索性研究 27第一節 情境與方法 27第二節 預測結果 30第六章 結論 35第一節 研究結果 35第二節 研究限制與未來方向 36參考文獻 37 zh_TW dc.format.extent 2109807 bytes - dc.format.mimetype application/pdf - dc.source.uri (資料來源) http://thesis.lib.nccu.edu.tw/record/#G0107356006 en_US dc.subject (關鍵詞) 電子零組件供應鏈 zh_TW dc.subject (關鍵詞) 需求預測 zh_TW dc.subject (關鍵詞) 機器學習 zh_TW dc.subject (關鍵詞) Semiconductor Supply chain en_US dc.subject (關鍵詞) Demand Forecast en_US dc.subject (關鍵詞) Machine Learning en_US dc.title (題名) 機器學習於高變異、不平衡的生產需求預測 zh_TW dc.title (題名) Machine Learning for Demand Forecasting with Highly Volatile and Imbalanced Data en_US dc.type (資料類型) thesis en_US dc.relation.reference (參考文獻) Breiman, L. (2001). Random forests. Machine Learning, vol.45, 5–32.Carbonneau, R., Laframboise, K., & Vahidov, R. (2008). Application of machine learning techniques for supply chain demand forecasting. European Journal of Operational Research, vol.184(3), 1140–1154.Chen, T., & Guestrin, C. (2016). XGBoost: A Scalable Tree Boosting System.Friedman, J. (2001). Greedy Function Approximation : A Gradient Boosting Machine. The Annals of Statistics, vol.29(5), 1189–1232.Hartigan, J. A., & Wong, M. A. (1979). A K-Means Clustering Algorithm. Journal of the Royal Statistical Society. Series C (Applied Statistics), vol.28(1), 100–108.Ma, S., & Fildes, R. (2020). Forecasting third-party mobile payments with implications for customer flow prediction. International Journal of Forecasting.Nikolopoulos, K., Syntetos, A. A., Boylan, J. E., Petropoulos, F., & Assimakopoulos, V. (2011). An aggregate-disaggregate intermittent demand approach (ADIDA) to forecasting: An empirical proposition and analysis. Journal of the OperationalQu, T., Zhang, J. H., Chan, F. T. S., Srivastava, R. S., Tiwari, M. K., & Park, W. Y. (2017). Demand prediction and price optimization for semi-luxury supermarket segment. Computers and Industrial Engineering, vol.113, 91–102.Rostami-Tabar, B., Babai, M. Z., Syntetos, A., & Ducq, Y. (2013). Demand Forecasting by Temporal Aggregation. Naval Research Logistics, vol.60(6), 479–Syntetos, A. A., Babai, Z., Boylan, J. E., Kolassa, S., & Nikolopoulos, K. (2016). Supply chain forecasting: Theory, practice, their gap and the future. European Journal of Operational Research, vol.252(1), 1–26.Syntetos, A. A., & Boylan, J. E. (2005). The accuracy of intermittent demand estimates. International Journal of Forecasting, vol.21(2), 303–314.Teunter, R. H., Syntetos, A. A., & Babai, M. Z. (2011). Intermittent demand: Linking forecasting to inventory obsolescence. European Journal of Operational Research, vol.214(3), 606–615.Vairagade, N., Logofatu, D., Muharemi, F., & Leon, F. (2019). Demand Forecasting Using Random Forest and Artificial Neural Network for Supply Chain Management. International Conference on Computational Collective Intelligence, 328–339.Willemain, T. R., Smart, C. N., Shockor, J. H., & DeSautels, P. A. (1994). Forecasting intermittent demand in manufacturing: a comparative evaluation of Croston’s method. International Journal of Forecasting, vol.10(4), 529–538. zh_TW dc.identifier.doi (DOI) 10.6814/NCCU202001637 en_US
