本研究以SAN聚合製程為研究對象，使用實廠數據建立類神經網路模型，並探討建模細節及模型應用。本研究除了利用數學軟體MATLAB®，建立SAN聚合製程之類神經網路模型外，並以程序模擬軟體Simulink®為平台設計模組，以實現模型功能，包括數值模擬、未來預測及情境模擬。
實廠苯乙烯丙烯腈樹脂聚合製程中，因缺少線上量測產品熔融指數的儀器，故欲建立模型作為Soft-sensor，模擬其數值。在過去的文獻中，因高分子製程複雜，建立理論模型難度較高，故選擇數據導向的黑箱模型方法，除較理論模型容易達成外，模型通常有不錯的擬合表現，其中，又以類神經網路方法為現今主流。
建模細節包含製程變數的分析及篩選、蒐集數據後的清理與補值、模型主要架構及細項的選用、數據分段及分類，以及相當關鍵的建模策略。本研究中依序提出了數種模型策略，最終的兩階段式模型結果顯示，引入之觀測變數若易受到設備影響而有差異，將使模型於設備更動時失去功用，並且需要重新蒐集數據及建模，非常不實際；一階段式模型則為了克服前述問題而提出，儘管模型的誤差仍稍有落後前者，但仍達到目標範圍內。
模型模組化後，已可同步連動實廠數據，除達成即時的數值模擬外，未來預測也利用了模型架構中的時間延遲設定來實現。情境模擬模組的設計，除了用於變數測試，再次檢視模型的合理性外，也用以測試品別轉換時輸出變數的響應，分別對四種品別轉換案例，找出最佳操作策略，並做為實廠調整的參考。

This study takes the styrene-acrylonitrile (SAN) copolymer process as the research object, and establish an Artificial neural network (ANN) model with the real-factory data. In this study, MATLAB® and Simulink® are used for the model training and the design of modules. And also, three functions, process monitoring, process prediction, and scenario based test are implemented in this study.
In the industrial SAN process, there is no online instrument for measuring the melt index (MI) of products. It is worth establishing a model as a soft-sensor for the variable estimation. In the past literature, it is harder to establish a theoretical model than black-box one due to the complexity of the polymerization processes. The black-box models are not only easier to achieve but also often have high-level performance. ANN method we chose in this study is one of famous ways to build black-box models.
From the results of the two-stage model, it is impractical to apply for intermediate variables estimation because the variables from equipment are usually changed after the process maintenance. Therefore, the one-stage model is the better approach that can overcome this issue. Although the performance of one-stage model is worse than former one, it still reach the acceptable range (MSE<0.05). All the three functions have developed for the end-user of SAN process. The results provide good reference values for the process operations.

[英文]
1. McCulloch, W. S. and Pitts, W. (1943). A Logical Calculus of the Idea Immanent in Nervous Activity. Bull. Math. Biol., Vol. 5, 115-133.
2. Levenberg, K. (1944). A Method for The Solution of Certain Non-linear Problems in Least Squares. Q. Appl. Math., Vol. 2, No. 2, pp. 164-168.
3. Marquardt, D. W. (1963). An Algorithm for Least-Squares Estimation of Nonlinear Parameters. J. Soc. Indust. Appl. Math. Vol. 2, pp. 431-441.
4. McClelland, T. L., Rumelhart, D. E. and the PDP Research Group (1986). Parallel Distributed Processing: Exploration in the Microstructure of Cognition. Vol. 1, The MIT Press., Cambridge.
5. Cybenko, G. (1989). Approximations by Superpositions of a Sigmoidal Function. Math. Control Signals Systems, Vol. 2, pp. 303-314.
6. Lin, T., Horne, B. G., Tino, P. and Giles C. L. (1996). Learning Long-term Dependences in NARX Recurrent Neural Networks. IEEE Trans. Neural Networks, Vol. 7, pp. 1329–1338.
7. Tamura, S. and Tateishi, M. (1997). Capabilities of a Four-Layered Feedforward Neural Network: Four Layers Versus Three. IEEE Trans. Neural Netw., Vol. 8, No. 2, pp. 251–255.
8. Zhang, J., Martin, E. B., Morris, A. J. and Kiparissides, C. (1997). Inferential Estimation of Polymer Quality Using Stacked Neural Networks. Comput. Chem. Eng. Vol. 21, Suppl., pp. S1025-S1030.
9. Zhang, J. (1999). Inferential Estimation of Polymer Quality Using Bootstrap Aggregated Neural Networks. Neural Netw., Vol. 12, pp. 927-938.
10. Fernandes, F. A. N. and Lona, L. M.F. (2005). Neural Network Applications in Polymerization Process. Braz. J. Chem. Eng., Vol. 22, No. 3, pp. 401-418.
11. Han, I.-S., Han, C. and Chung, C.-B. (2005). Melt Index Modeling with Support Vector Machines, Partial Least Squares, and Artificial Neural Networks. J. Appl. Polym. Sci., Vol. 95, pp. 967–974.
12. Zhang, J., Jin, Q. and Xu, Y. (2006). Inferential Estimation of Polymer Melt Index Using Sequentially Trained Bootstrap Aggregated Neural Networks. Chem. Eng. Technol. Vol. 29, No. 4, pp. 442-448.
13. Shi, J., Liu, X. and Sun, Y. (2006). Melt Index Prediction by Neural Networks Based on Independent Component Analysis and Multi-Scale Analysis. Neurocomputing, Vol. 70, pp. 280-287.
14. Xia, L. and Pan, H. (2010). Inferential Estimation of Polypropylene Melt Index Using Stacked Neural Networks Based on Absolute Error Criteria. Paper presented at the International Conference on Computer, Mechatronics, Control and Electronic Engineering(CMCE), Changchun, China.
15. Xia, L., Pan, H., Zhou, M., Cai, Y. and Sun, X. (2011). Modeling Methodology Based on Stacked Neural Networks for Inferential Prediction of Propylene Melt Index. Appl. Mech. Mater., Vol. 55-57, pp. 670-674.
16. Lou, H., Su, H., Xie L., Gu, Y. and Rong, G. (2012). Inferential Model for Industrial Propylene Melt Index Prediction with Embedded Priori Knowledge and Delay Estimation. Ind. Eng. Chem. Res., Vol. 51, pp. 8510-8525.
17. Jumari, N. F. and Yusof, K. M. (2017). Comparison of Melt Flow Index of Propylene Polymerization in Loop Reactors Using First Principles and Artificial Neural Network models. Chem. Eng. Trans., Vol. 56, pp.163-168.
18. Liu, Y., Yang, C., Gao, Z. and Yao, Y. (2018). Ensemble Deep Kernel Learning with Application to Quality Prediction in industrial Polymerization Processes. Chemom. Intell. Lab. Syst., Vol. 174, pp. 15-21.

[中文]
1. 余宗興，類神經網路模式在高壓共聚合製程上的應用，碩士論文，台灣科技大學，2006年。
2. 楊宗樺，類神經網路之動態估測器於高壓共聚合反應之應用，碩士論文，台灣大學，2007年。