Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/93174| Title: | 適用於物聯網機器服務業的深度學習輔助容量管理 Deep Learning Assisted Capacity Management for IoT-Enabled Services Industries |
| Authors: | 曹竣瑋 Chun-Wei Tsao |
| Advisor: | 陳靜枝 Ching-Chin Chern |
| Keyword: | 物聯網機器,大數據,時間序列分析,深度學習, IoT machine data,Big Data,Autoencoder,Deep Learning, |
| Publication Year : | 2024 |
| Degree: | 碩士 |
| Abstract: | 近年來,因為物聯網 (IoT, Internet of Things) 機器的出現,配備機器的服務業在收集資料的方式可以轉為自動化。除了可以精確的得到機器運轉的情況,也能過濾掉機器使用狀況異常的訊號。從物聯網感測器所收集到的資料具備大數據的性質,包含資料量(Volume),以及速度(Velocity),和多樣性(Variety)。這樣的資料性質可以針對特定機器的運轉情況作出客製化的處理,詳細記錄時間的資訊也能更有效地執行時間序列分析。
本研究使用洗衣租賃業所提供的物聯網機器資料,針對700多台洗衣機總共兩年的資料進行時間序列分析及預測,並期望幫助服務業避免使用經驗做出新增機器的決策,轉向使用資料輔助預測的方式達到資源分配的目的。我們首先利用物聯網機器狀態轉變的訊號進行處理,依照不同的時間段切分運轉情況。再針對所有的機器使用狀況建立深度學習模型,搭自編碼器(Autoencoder)技術處理複雜大量的物聯網資料,進行準確的時間序列預測,可以得出特定機器未來的運轉狀況。除此之外,我們提出兩階段的深度學習架構,使用分群演算法搭配分類演算法,可以使用所收集到的物聯網機器資料建立模型,幫助沒有歷史資料的機器,針對機器的性質,所要放置的地點進行預測,計算出未來可能的運轉情況。 根據本研究所提出的深度學習預測架構,企業可以針對物聯網機器所預測出的運轉情況和當前的資源分配做出比較,了解機器分配的情況並做出調整。不管特定機器是否有歷史資料,都能夠運用我們提出的架構,得到準確的運轉情況預測。 In recent years, the advent of Internet of Things (IoT) devices has revolutionized data collection in the service industry, enabling automation and precise monitoring of machine operations. IoT sensors generate large-scale data characterized by volume, velocity, and variety, which can be tailored to analyze specific machine performance and facilitate effective time series analysis. This study leverages IoT data from a laundromat industry, analyzing over two years of data from more than 700 washing machines. The goal is to transition from experience-based decision-making to data-driven predictions for resource allocation. We process signals from IoT devices to segment machine operations over different temporal dependencies. Using deep learning models and autoencoder technique, we accurately forecast time series data to predict future machine performance. Additionally, we propose a two-phase deep learning framework that combines clustering and classification algorithms. This approach allows us to build predictive models for machines without historical data, estimating future performance based on related features such as machine characteristics and deployment locations. Our predictive framework enables businesses to compare forecasted machine operations with current resource allocation, optimizing machine distribution and adjustments. Whether or not a machine has historical data, our approach provides accurate performance predictions. |
| URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/93174 |
| DOI: | 10.6342/NTU202401885 |
| Fulltext Rights: | 同意授權(全球公開) |
| Appears in Collections: | 資訊管理學系 |
Files in This Item:
| File | Size | Format | |
|---|---|---|---|
| ntu-112-2.pdf | 2.16 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.
