臺灣博碩士論文加值系統

因應的彈性製造發展及節能減碳的趨勢，將用電預測運用於工廠中成為重要發展趨勢，在理論上此維護方式突破了傳統用電分析方式，透過深度學習模型學習力使用電情況，提高工廠用電的負載率，減少工廠用電超約情況，且同時提供工單生產時用電評估，是具有顯著的經濟效益。
本文針對工廠提出用電設備能耗預測模式建構，工廠能耗以辦公設備與生產設備為主要來源，因此本文研究中對辦公設備與生產設備的能耗，藉由分析天氣與生產工單觀察與設備間的關係進行探討。
研究方法分為四個步驟進行 : 第一步蒐集工廠設備運行中的相關資料；第二步將針對蒐集的工廠運行資料，進行資料預處理與資料融合，以利模型訓練與預測；第三步使用深度學習演算法的方式對工廠用電進行未來情況預測，關於深度學習演算法本文採用四種深度學習模型，並針對時間序列模型比較時間長度對模型預測的影響，而後使用四種模型評估方式來對預測模型進行評估，選擇出最佳預測模型對工廠用電進行用電趨勢，經分析評估後，顯示混和模型(CNN-LSTM)在時間長度30時為最適合之預測模型，因此利用CNN-LSTM對工廠進行預測未來趨勢，CNN-LSTM模型預測工廠資料的均方根誤差、絕對平均誤差、絕對平均百分比及決定係數(R Squared)分別為2.7(KW2)、24.2(KW)、0.63(%)及0.92；第四步驟依據預測結果對工廠用電進行預測診斷以及工單評估。
總之，用電預測模式是一種整合工單系統及用電預測與評估的綜合技術，比傳統用電警示更加卓越的方式，具有用電警示、預測和用電分析技術的優點，此技術可利用多感測器監控，藉由各種資料收集技術和深度學習模型預測和評估工廠的用電情況，經由數據監控和分析，預測用電量情況給予工單安排建議。

The prediction in manufacturing industry is an important trend for flexible manufacturing system (FMS), saving energy and carbon reduction. In theory, maintains method of this paper breach the traditional method to analyze energy usage. It can improve rate of loading of energy in factory by using deep learning method to predict. It can reduce over-used energy and provide an advice for manufacture order at energy.
This paper builds an architecture for predict energy consumption in factory. In factory, energy consumption has two parts. One is office equipment and other is production line. Therefore, the research is directed at energy consumption in offices and production lines and discuss by analyzing the relationship between weather, production work order observation and equipment.
The research method has four steps. First, collecting data in a factory about the usage of the office and information about production. Second, preprocessing the collected data from the factory and merging the data from the office, production line and weather for module training and predicting. Third, using the deep learning module as a forecasting module to predict the consumption in the factory. The thesis uses five deep learning modules and compared each module at a different time length. The thesis uses four estimated methods to estimate the performance of module forecasting and will choose one module that has the best performance to predict consumption of manufacturers. After analyzing each variable, the results show the hybrid module (CNN-LSTM) at time length 30 has the best performance in forecasting energy consumption manufacturers. Therefore, the thesis recommends CNN-LSTM as the module to predict the direction of energy consumption in manufacturers. The module CNN-LSTM predicts the result on Root Mean Squared Error, Absolute Mean Error, Percentage Absolute Mean Error and R Squared is 2.7KW2, 24.2KW, 0.63% and 0.92. Last, the thesis uses estimated results to choose a module to predict the energy consumption in manufacturing.

摘要 ........................i
Abstract.........................iii
誌謝 .........................v
目錄 .........................vi
表目錄 .........................viii
圖目錄 .........................ix
符號 .........................xi
第一章 緒論.....................1
1.1 研究背景與動機 .........1
1.2 研究方法 .................2
1.3 研究流程 .................3
1.4 論文章節 .................4
第二章 文獻探討 .................5
2.1 彈性製造 .................5
2.2 電力 .................6
2.3 用電預測 .................7
2.3.1 容量警示系統 .........8
2.3.2 預測場合 .................8
2.4 深度學習模型 .........9
2.4.1 回歸法 .................9
2.4.1.1 多層感知器 ........10
2.4.1.2 卷積神經網路 ........12
2.4.2 時間序列法 ........16
2.4.2.1 長時間記憶神經網路 ......16
2.4.2.2 注意力機制 ........18
2.4.2.3 混合模型 ................20
2.5 本章小節 ................21
第三章 電力模型預測模式建 ......22
3.1 資料預處理 ........23
3.2 Min-Max正規化 ........23
3.3 資料融合 ................24
3.4 深度學習預測 ........25
3.4.1 多層感知器 ........25
3.4.2 卷積神經網路 ........26
3.4.3 長短期記憶神經網路 .....27
3.4.4 LSTM注意力機制 ........28
3.4.5 混合模型 ................29
第四章 電力模型預測建構實作 ....31
4.1 Python ................31
4.2 Keras ................31
4.3 模擬用電運行 ........32
4.4 資料預處理 ........35
4.5 資料融合 ................39
4.6 深度學習預測模型 ........40
4.6.1 多層感知器 ........41
4.6.2 CNN ................42
4.6.3 LSTM-NN ................44
4.6.4 注意力機制 ........47
4.6.5 混合模型 ................48
4.7 模型評估 ................51
第五章 模型驗證與討論 ........53
5.1 驗證一 ................54
5.2 驗證二 ................56
結論 ........................59
參考文獻 ........................60
Extended Abstract 64

[1]Aniekan Essien, Cinzia Giannetti, 2020, A Deep Learning Model for Smart Manufacturing Using Convolutional LSTM Neural Network Autoencoders
[2]Ijax Ul Haq, et al, 2021, Learning-Based Energy Consumption Prediction Model for Residential and Commercial Sectors
[3]Chibuzor N Obiora and Ahme Ali, 2021, Effective Implementation of Convolutional Long Short-Term Memory(ConvLSTM) Network in Forecasting Solar Irradiance
[4]Fazli Wahid, et al, 2017, Prediction of Energy Consumption in the Buildings Using Multi-Layer Perceptron and Random Forest.
[5]Y. Takama and D. Ursino, 2013, A Review of Data Fusion Techniques
[6]Dana Lahat, Tülay Adah, Christian Jutten, 2015, Multimodal Data Fusion: An Overview of Methods, Challenges and Prospects
[7]Ralf C. Staudemeyer and Eric Rothstein Morris, 2019, Understanding LSTM:a tutorial into Long Short-Term Memory Recurrent Neurral Networks
[8]Sepp Hochreiter and Jürgen Schmidhuber, 1997, LONG SHORT-TERM MEMORY
[9]Valentina Emilla Balas and Nikos E. Mastorakis, 2009, Multilayer perceptron and neural networks
[10]Xuan Zheng, 2019, AT-LSTM:An Attention-based LSTM Model for Financial Time Series Prediction
[11]Loannis E. Livieris, Emmanuel Pintelas and Panagiotis Pintelas, 2020, A CNN-LSTM model for gold price time-series forecasting
[12]Rikiya Yamashita, et al, 2018, Convolutional neural networks: an overview and application in radiology
[13]Jessica Walther and Matthias Weigold, 2021, A Systematic Review on Predicting and Forecasting the Electrical Energy Consumption in the Manufacturing Industry
[14]Gökan MAY, et al, 2017, Energy management in manufacturing:From literature review to a conceptual framework
[15]Ali Agga, et al, 2021, Short-term self consumption PV plant power production forecasts based on hybrid CNN-LSTM, ConvLSTM models
[16]NECATİ DENİZ YÜCEL, 2005, Simulation of a Flexible Manufacturing Sytem: A Pilot Implementation
[17]M. T. Alvela Nieto, et al, 2019, Enabling Energy Efficiency in Manufacturing Environments Through Deep Learning Approaches: Leassons Learned
[18]Nour-Eddin El Faouzi, et al, 2016, Data Fusionn for ITS: Techniques and Research Needs
[19]Zhifeng Lin, Lianglun Cheng and Guoheng Huang, 2019, Electricity Consumption Prediction Based on LSTM with Attention Mechanism
[20]Tuong Le, et al, 2019, Improving Electric Energy Consumption Prediction Using CNN and Bi-LSTM
[21]S. Y. Kung, 2022, XNAS: A Regressive/Progressive NAS for Deep Learning
[22]George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel, Greta M. Ljung, 2015, Time Series Analysis:Forecasting and Control, Edition
[23]Xiaoxian Yang, et al, 2019, Novel Financial Capital Flow Forecast Framework Using Time Series Theory and Deep Learning: A Case Study Analysis of Yu’e Bao Transaction Data
[24]Tiago Santos, Roman Kern, 2016, A Literature Survey of Early Time Series Classification and Deep Learning
[25]Alwin Gerolt Oerlemans, 1992, Production Planning for Flexible Manufacturing Systems
[26]Anshuman Tripathy, 2013, Machine Loading in a Flexible Manufacturing System
[27]Besim Bilalli, 2018, Learning the Impact of Data Pre-processing in Data Analysis
[28]Ivan Miguel Pires, et al, 2020, Homogeneous Data Normalization and Deep Learning: A Case Study in Human Activity Classification
[29]Anish Hemant Narkhede, 2019, Human Activity Recognition Based on Multimodal Body Sensing
[30]Sammy Siu, 1990, Non-Linear Adaptive Equalization Based on a Multi-Layer Perceptron Architecture
[31]Anthony L. Caterini, 2017, A Novel Mathematical Framework for the Analysis of Neural Networks
[32]Tianzhe Bao, Shengquan Xie and Zhi-Qiang Zhang, 2021, A CNN-LSTM Hybrid Model for Wrist Kinematics Estimation Using Surface Electromyography
[33]Yanliang Jin, Dijia Wu and Weisi Guo, 2020, Attention-Based LSTM with Filter Mechanism for Entity Relation Classification
[34]台灣電力公司, 2022,電價表
[35]台灣電力公司, 2015, 台灣含氟溫室氣體排放減量報導第17期