化學工廠平時運作，最重要除了操作安全之外，產品品質及成本將會影響客戶採購的意願。以往確認品質需要取樣至實驗室，使用價格昂貴的分析儀器分析，或安裝線上分析儀器，以便能了解品質狀況，做必要之操作調整。因等候或分析儀器經長年累月運作性能會衰退，錯失調整時機及導致分析失準，因應此等議題已有商業產品以軟體儀表進行量測，並同時與實體分析儀分析來比對，或直接替代實體分析儀，提升操作效率。
本研究以Tennessee Eastman Process資料為基礎，研究正常製程及不正常製程時，使用製程及操作變數，進行軟體儀表之建模。使用統計方法：多變數線性迴歸、二次效應曲面法模型。神經網路方法：神經網路、自動編譯器神經網路、長短記憶模型遞迴神經網路。在各種之數據量下，進行產品品質軟體儀表之建模預測，並與實體分析儀進行比較。
結果顯示，正常製程在數據量少的情況下，模型對訓練組的誤差雖低，但對測試組的結果不好，此種狀況猶如實驗室分析，提供的資料正常但不足，此時統計學習所產生簡單的模型，即有不錯的效果。當數據量多且密集時，使用神經網路方法：長短記憶模型遞迴神經網路建模，對測試組會有比較好的預測效果。不正常製程，普遍因為資料變動大，不論是統計方法或神經網路方法，均有不錯的效果，但應用上仍要注意資料的合理性。

Chemical plants operate in a normal state. The most important issue is safety. Other than this, the customer will determine with quality and cost whether they will buy or not. To check the quality, we need to do the sampling to experiment. Then using expensive instruments or install an online analyzer to analyze. Depend on the purity to do some adjusting. Due to waiting or the analyzer will be decay after a long time operation. The operator will lose the timing to adjust, and the accuracy will be lost. To facing such problems, for business applications, some products get quality from soft sensors. It could check with a rugged analyzer at the same time or replace the hardware to increase operational efficiency.
This research uses the Tennessee Eastman Process database. Study in normal process and abnormal process. To build the models by the process and operation variables will be used. For statics, methods are multiple linear regression, quadratic response surface method. For neural networks, methods are the artificial neural network, autoencoder neural network, and long-short terms memory recurrent neural network. Training the model using different data sizes and then compare it with accurate data.
The results are well for the normal process when the sample data size is small, even for the model error small for the training set. However, the accuracy set is not suitable for the testing set. This situation, like the data, comes from experiments, and the data is insufficient. For this case, the statics model is better. When the data size increases, the neural network method like long-short term memory recurrent neural network will be good. Due to the data variance significance, both the statics and neural network methods will be suitable for abnormal processes. For an actual application, it is must check data sourcing reasonably.

中文摘要 ii
英文摘要 iii
誌謝 iv
目錄 v
圖目錄 List of Figure vii
表目錄 List of Table viii
符號說明 ix
第一章 緒論 1
1.1 研究背景與重要性 1
1.1.1 前言 1
1.1.2 模擬及人工智慧 1
1.2 研究動機 4
1.3 研究目的 4
1.4 論文結構 5
第二章 文獻回顧 6
第三章 系統架構 16
3.1 資料檢查 18
3.2 建模預測方法 19
3.2.1 多變數線性迴歸模型 MLR 19
3.2.2 二次效應曲面法模型 QRS 19
3.2.3 神經網路模型 DNN 19
3.2.4 自動編譯神經網路模型 AE 19
3.2.5 長短記憶模型遞迴神經網路 LSTM-RNN 20
3.3 比較基準 22
3.4 TEP資料集說明 25
3.5 軟體使用 26
第四章 實證研究 28
4.1 正常製程 31
4.2 不正常製程 41
第五章 結論 51
5.1 研究貢獻 51
5.2 未來研究方向 51
參考文獻 52
附錄一 正常時模型ytrue vs ypred之X-Y Chart 56
附錄二 不正常時模型ytrue vs ypred之X-Y Chart 58

[1] 汪上曉, et al., “化學工程學系課程與教學在人工智慧發展下的因應與演進,” 化工 67(2): 98-103, 2020.
[2] 簡禎富 “工業3.5 : 台灣企業邁向智慧製造與數位決策的戰略,” 天下, 2019.
[3] F. Yu, et al., “Feature based causality analysis and its applications in soft sensor modeling,” IFAC PaterOnLine 53(2) 138-143, 2020.
[4] E. Örs, et al., “A Conceptual Framework for AI-based Operational Digital Twin in Chemical Process Engineering.” 2020 IEEE International Conference on Engineering, Technology, and Innovation (ICE/ITMC), 2020.
[5] H. H. Nguyen, et al., “Towards nominal stability certification of deep learning-based controllers,” 2020 American Control Conference (ACC), 2020.
[6] J. G. Wang and J. H. Zhao, et al. “Deep learning based automatic maintenance of soft sensors used in wastewater treatment plants,” 2020 IEEE-HYDCON, 2020.
[7] L. Fang, et al., “Design and Development of the AI-assisted Safety System for Hazardous Plant,” 2020 13th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI), 2020.
[8] Y. Cong, et al., “Multirate Dynamic Process Monitoring Based on Multirate Linear Gaussian State-Space Model,” IEEE Transactions on Automation Science and Engineering 16(4): 1708-1719, 2019.
[9] R. Borrison, et al. “Data Preparation for Data Mining in Chemical Plants using Big Data,” 2019 IEEE 17th International Conference on Industrial Informatics (INDIN), 2019.
[10] S. O. Baarimah, et al., “Using Chemical Composition of Crude Oil and Artificial Intelligence Techniques to Predict the Reservoir Fluid Properties,” 2019 First International Conference of Intelligent Computing and Engineering (ICOICE), 2019.
[11] W. Wang, et al., “Detection of data injection attack in industrial control system using long short term memory recurrent neural network,” 2018 13th IEEE Conference on Industrial Electronics and Applications (ICIEA), 2018.
[12] L. A. Renforth, et al., “A Smart Condition Monitoring System for HV Networks with Artificial Intelligence, Augmented Reality and Virtual Reality,” Copyright Material IEEE, Paper No. PCIC-2018-37. 2018 IEEE Petroleum and Chemical Industry Technical Conference (PCIC), 2018.
[13] X. Zhang and M. Kano, “Soft-sensor reliability evaluation and y-analyzer fault identification with applications to vinyl acetate monomer (VAM) benchmark process,” 2017 56th Annual Conference of the Society of Instrument and Control Engineers of Japan (SICE), 2017.
[14] M. A. Taube et al., “Cost-competitive “soft sensor” for determining product recovery in industrial methanol,” 2017 Moratuwa Engineering Research Conference (MERCon), 2017.
[15] F. Luo, et al., “Soft-sensing modeling based on GK-LSSVM method for online predictions of BOD in activated sludge process,” 2017 International Conference on Robotics and Automation Sciences (ICRAS), 2017.
[16] J. G. Wang, et al., “Deep learning-based soft-sensing method for operation optimization of coke dry quenching process,” 2016 35th Chinese Control Conference (CCC), 2016.
[17] Z. Marinković, et al., ”A neural network approach for safety monitoring applications,” 2016 IEEE Sensors Applications Symposium (SAS), 2016.
[18] J. Kortela and S. L. Jämsä-Jounela, “Improvement of load-following capacity of grate boilers based on the combustion power soft-sensor,” 2016 IEEE 21st International Conference on Emerging Technologies and Factory Automation (ETFA), 2016.
[19] F. Belkhir and G. Frey, “Model-driven soft sensor for predicting biomass calorific value in combustion power plants,” 2016 IEEE 11th Conference on Industrial Electronics and Applications (ICIEA), 2016.
[20] A. K. Pani and H. K. Mohanta, “A hybrid soft sensing approach of a cement mill using principal component analysis and artificial neural networks,” 2013 3rd IEEE International Advance Computing Conference (IACC), 2013.
[21] Y. Lei and H. Yang, “A Gaussian process ensemble modeling method based on boosting algorithm,” Proceedings of the 32nd Chinese Control Conference, 2013.
[22] R. Grbic, et al., “Adaptive soft sensor for online prediction and process monitoring based on a mixture of Gaussian process models,” Computers and Chemical Engineering 58 84-97, 2013.
[23] Y. Wang, et al., “A soft sensor for carbon content of spent catalyst in a continuous reforming plant using LSSVM-GA,” Proceedings of the 31st Chinese Control Conference, 2012.
[24] A. K. Pani, et al., ”Data driven soft sensor of a cement mill using generalized regression neural network,” 2012 International Conference on Data Science & Engineering (ICDSE), 2012.
[25] S. Nazari, et al., “An improved algebraic geometric solution to the identification of switched ARX models with noise,” Proceedings of the 2011 American Control Conference, 2011.
[26] J. Kim, et al., “Modeling of Chemical Plant’s Rectifying Towers Using Artificial Neural Networks,” 2011 Third International Conference on Computational Intelligence, Communication Systems and Networks, 2011.
[27] J. Chen, et al., “Application of least squares support vector machine in soft sensor of traditional Chinese Medicine extraction,” 2011 Seventh International Conference on Natural Computation. 2011.
[28] M. Xinglu, “Research on Soft Sensing Technology for Column Efficiency of Ion Chromatography,” 2010 Third International Symposium on Information Processing, 2010.
[29] L. Xia and H. Pan, “Inferential estimation of polypropylene melt index using stacked neural networks based on absolute error criteria,” 2010 International Conference on Computer, Mechatronics, Control and Electronic Engineering, 2010.
[30] H. Tian, et al. “Soft sensor for polypropylene melt index based on improved orthogonal least squares,” 2010 8th World Congress on Intelligent Control and Automation, 2010.
[31] S. Graziani, et al. (2010). “Soft sensor for a Propylene Splitter with seasonal variations,” 2010 IEEE Instrumentation & Measurement Technology Conference Proceedings,2010.
[32] M. Nakaya, et al., “A new estimation method by utilizing online tracking simulator,” 2009 ICCAS-SICE, 2009
[33] L. Zhu, et al. “Soft sensor for distillation column feeds,” 2009 Chinese Control and Decision Conference, 2009.
[34] A. Di Bella, et al. “Selection of regressors using correlation analysis to design a Virtual Instrument for an SRU of a refinery,” 2007 Mediterranean Conference on Control & Automation, 2007.
[35] X. Jin, et al., “Development of PP Melt Index Inferential Model,” 2006 6th World Congress on Intelligent Control and Automation, 2006.
[36] H. Q. Wang, et al. “NN-PLS based online component soft-analyzer for Tennessee Eastman process,” Fifth World Congress on Intelligent Control and Automation (IEEE Cat. No.04EX788), 2004.
[37] S. Li and F. Qian, “Application of soft sensors in the estimation of ethylene distillation column composition, ”Fifth World Congress on Intelligent Control and Automation (IEEE Cat. No.04EX788), 2004.
[38] D. J. H. Wilson and G. W. Irwin, “PLS modelling and fault detection on the Tennessee Eastman benchmark,” Proceedings of the 1999 American Control Conference (Cat. No. 99CH36251), 1999.
[39] D. Ruiz, et al.” Artificial neural networks applied to online fault diagnosis in chemical plants,” 1999 7th IEEE International Conference on Emerging Technologies and Factory Automation. Proceedings ETFA ‘99 (Cat. No.99TH8467), 1999.
[40] S. Parthasarathy, et al., “Prediction of flooding in an absorption column using neural networks,” Proceedings of the 1999 IEEE International Conference on Control Applications (Cat. No.99CH36328), 1999.
[41] D. J. H. Wilson and G. W. Irwin, ”Multivariate SPC using radial basis functions,” UKACC International Conference on Control ‘98 (Conf. Publ. No. 455), 1998.
[42] D. Saez, et al. “Prediction of water chemical properties in the cycle of a coal power plant using artificial neural networks,” 1998 IEEE International Joint Conference on Neural Networks Proceedings. IEEE World Congress on Computational Intelligence (Cat. No.98CH36227), 1998.
[43] D. J. H. Wilson, et al. “Neural networks and multivariate SPC,” IEE Colloquium on Fault Diagnosis in Process Systems, 1997.
[44] Y. L. Rouzic, et al. “Soft sensor for adaptive PH control, an industrial application,” 1997 European Control Conference (ECC), 1997.
[45] M. J. Willis “Soft-sensing via artificial neural networks,” IEE Colloquium on Automation and Control in Food Processing, 1992.