為確保電廠的穩定運轉，運轉員需一直監控電廠狀態，當肇始事件發生時，大量的電廠量測訊號會有劇烈變化，於此同時運轉員需從這些訊號中判斷出事件類別以便進行對應的處理。本論文提出一個基於電廠模擬器之模擬數據的肇始事件偵測與辨識系統，用於輔助運轉員快速判斷事件類別並妥善的處理，以避免更嚴重的損害發生。
在偵測系統中，我們透過分析剛擷取到的訊號與預估的穩定運轉訊號間是否有過大的差異來斷定肇始事件發生與否。我們分別運用Hotelling’s T2檢定以及長短期記憶來達成事件偵測，其中Hotelling’s T2檢定因無法消除穩定運轉訊號內的非穩態趨勢，導致常有提前偵測的發生；長短期記憶則可透過非線性預測來消除此問題的影響，故大部分的事件偵測可被準確偵測。此兩種方法皆採用閾值來進行狀態判斷，當誤差持續超過閾值10秒時，便被判定為肇始事件。
我們的辨識系統則透過四種不同的特徵萃取演算法來擷取得事件的特徵，再透過最近鄰居分類法來進行事件辨識。所採用的特徵萃取演算法包含：時間積分法、離散小波轉換、主成分分析以及感測器區塊投影法，其中感測器區塊投影因同時考慮訊號彼此之間的關聯性，故在辨識上有高達98.35%的辨識準確率。另外，事件辨識的準確率會隨著用來萃取訊號特徵的訊號時間長度變長而變低，其原因是當某事件發生時訊號會有劇烈的波動，同時訊號中也會含有該事件獨特的特徵，但當時間越久，訊號就會趨於穩定並為零，此時不同事件間訊號的差異就會變小，也導致辨識準確率的下降。

To ensure the safe operation of nuclear power plants (NPPs), being able to prevent the initiating events (IEs) from escalating into severe accidents becomes essential. Although various methods have been presented to detect and identify IEs, their development relies on the data simulated by the computer codes like MAAP. This study proposes an on-site simulator data-based system comprising event detection and identification to help operators promptly discriminate IEs and prevent the situation from worsening. When an IE occurs, changes in the sensing signals will happen, which allow an abnormal state to be discriminated from the normal operation. Hotelling’s T2 test and long short-term memory (LSTM) are the approaches implemented as the detection mechanism. To distinguish normal and abnormal states, thresholds for Hotelling’s T2 statistic under normal operation and interquartile range (IQR)-based outlier detection method for LSTM are required. The alarm will not be triggered unless the discrepancy continues to exceed the thresholds for 10 s. Owing to the unremoved fluctuations, Hotelling’s T2 test always detects the events early, while the LSTM-based detector detects most events accurately since it follows the trends well. After the events are detected, different durations of signals are considered as the event data. Four feature extractors are applied to find the low-dimensional representations for IEs that will be stored in the event database or sent to the classifier to identify the event type. Among the extractors, sensor type-wised block projection attained the highest identification rate of 98.35 % with the 5 s data duration since it is the only method that retains the interrelations among sensors of the same type while extracting features. Also, the readings in the sensors decline and flatten over time and cause the identification rates to decrease as the data duration increases.

摘要 i
Abstract ii
致謝 iii
目錄 v
表目錄 vii
圖目錄 viii
第一章 緒論 1
1.1 研究緣起 1
1.2 文獻回顧 2
1.3 研究方法與流程 5
第二章 偵測系統 7
2.1 基本假設 7
2.2 Hotelling’s T2檢定 7
2.3 長短期記憶 9
2.3.1 深度學習 9
2.3.2 長短期記憶 11
第三章 辨識系統 15
3.1 特徵萃取 15
3.1.1 時間積分法 15
摘要 i
Abstract ii
致謝 iii
目錄 v
表目錄 vii
圖目錄 viii
第一章 緒論 1
1.1 研究緣起 1
1.2 文獻回顧 2
1.3 研究方法與流程 5
第二章 偵測系統 7
2.1 基本假設 7
2.2 Hotelling’s T2檢定 7
2.3 長短期記憶 9
2.3.1 深度學習 9
2.3.2 長短期記憶 11
第三章 辨識系統 15
3.1 特徵萃取 15
3.1.1 時間積分法 15
3.1.2 離散小波轉換 16
3.1.3 主成分分析 16
3.1.4 感測器區塊投影法 17
3.2 最近鄰居分類法 18
第四章 實驗結果與討論 20
4.1 數據模擬 20
4.1.1 模擬器 20
4.1.2 事件模擬與設定 21
4.1.3 數據前處理 24
4.2 偵測結果與討論 24
4.3 辨識結果與討論 29
第五章 結論 33
參考文獻 35

[1] M. M. Abu-Khader, "Recent advances in nuclear power: A review," Progress in Nuclear Energy, vol. 51, no. 2, pp. 225-235, 2009.
[2] J. M. Broughton, P. Kuan, D. A. Petti, and E. Tolman, "A scenario of the Three Mile Island unit 2 accident," Nuclear Technology, vol. 87, no. 1, pp. 34-53, 1989.
[3] C. Lu et al., "Nuclear power plants with artificial intelligence in industry 4.0 era: Top-level design and current applications—A systemic review," IEEE Access, vol. 8, pp. 194315-194332, 2020.
[4] H. M. Hashemian, "On-line monitoring applications in nuclear power plants," Progress in Nuclear Energy, vol. 53, no. 2, pp. 167-181, 2011.
[5] J. Ma and J. Jiang, "Applications of fault detection and diagnosis methods in nuclear power plants: A review," Progress in nuclear energy, vol. 53, no. 3, pp. 255-266, 2011.
[6] V. Chandola, A. Banerjee, and V. Kumar, "Anomaly detection: A survey," ACM computing surveys (CSUR), vol. 41, no. 3, pp. 1-58, 2009.
[7] F. Giannoni, M. Mancini, and F. Marinelli, "Anomaly detection models for IoT time series data," arXiv preprint arXiv:1812.00890, 2018.
[8] I. S. Kim, "Computerized systems for on-line management of failures: a state-of-the-art discussion of alarm systems and diagnostic systems applied in the nuclear industry," Reliability Engineering & System Safety, vol. 44, no. 3, pp. 279-295, 1994.
[9] K. C. Kwon, J. H. Kim, and P. H. Seong, "Hidden Markov model‐based real‐time transient identifications in nuclear power plants," International journal of intelligent systems, vol. 17, no. 8, pp. 791-811, 2002.
[10] M. G. Na, W. S. Park, and D. H. Lim, "Detection and diagnostics of loss of coolant accidents using support vector machines," IEEE Transactions on Nuclear Science, vol. 55, no. 1, pp. 628-636, 2008.
[11] M. M. Breunig, H.-P. Kriegel, R. T. Ng, and J. Sander, "LOF: identifying density-based local outliers," in Proceedings of the 2000 ACM SIGMOD international conference on Management of data, 2000, pp. 93-104.
[12] J. Zhang and H. Wang, "Detecting outlying subspaces for high-dimensional data: the new task, algorithms, and performance," Knowledge and information systems, vol. 10, no. 3, pp. 333-355, 2006.
[13] O. Kramer, "K-nearest neighbors," in Dimensionality reduction with unsupervised nearest neighbors: Springer, 2013, pp. 13-23.
[14] K. Moshkbar-Bakhshayesh and M. B. Ghofrani, "Transient identification in nuclear power plants: A review," Progress in Nuclear Energy, vol. 67, pp. 23-32, 2013.
[15] Z. Gao, C. Cecati, and S. X. Ding, "A survey of fault diagnosis and fault-tolerant techniques—Part I: Fault diagnosis with model-based and signal-based approaches," IEEE transactions on industrial electronics, vol. 62, no. 6, pp. 3757-3767, 2015.
[16] J. Ma and J. Jiang, "Applications of fault diagnosis in nuclear power plants: an introductory survey," IFAC Proceedings Volumes, vol. 42, no. 8, pp. 1150-1161, 2009.
[17] R. Li and J. H. Olson, "Fault detection and diagnosis in a closed-loop nonlinear distillation process: Application of extended Kalman filters," Industrial & Engineering Chemistry Research, vol. 30, no. 5, pp. 898-908, 1991.
[18] H. Liu, D. Liu, C. Lu, and X. Wang, "FAULT DIAGNOSIS OF HYDRAULIC SERVO SYSTEM USING THE UNSCENTED K ALMAN FILTER," Asian Journal of Control, vol. 16, no. 6, pp. 1713-1725, 2014.
[19] T.-Y. Hsiao, C. Lin, and Y. Yuann, "Identification of initiating events for pressurized water reactor accidents," Annals of Nuclear Energy, vol. 37, no. 11, pp. 1502-1512, 2010.
[20] D. Roverso, "Plant diagnostics by transient classification: The aladdin approach," International Journal of Intelligent Systems, vol. 17, no. 8, pp. 767-790, 2002.
[21] B. Lu and B. Upadhyaya, "Monitoring and fault diagnosis of the steam generator system of a nuclear power plant using data-driven modeling and residual space analysis," Annals of Nuclear Energy, vol. 32, no. 9, pp. 897-912, 2005.
[22] N. Zavaljevski and K. C. Gross, "Sensor fault detection in nuclear power plants using multivariate state estimation technique and support vector machines," Argonne National Lab., Argonne, IL (US), 2000.
[23] T. Santosh, A. Srivastava, V. S. Rao, A. Ghosh, and H. Kushwaha, "Diagnostic system for identification of accident scenarios in nuclear power plants using artificial neural networks," Reliability Engineering & System Safety, vol. 94, no. 3, pp. 759-762, 2009.
[24] K. Mo, S. J. Lee, and P. H. Seong, "A dynamic neural network aggregation model for transient diagnosis in nuclear power plants," Progress in nuclear energy, vol. 49, no. 3, pp. 262-272, 2007.
[25] T.-H. Lin, T.-C. Wang, and S.-C. Wu, "Deep learning schemes for event identification and signal reconstruction in nuclear power plants with sensor faults," Annals of Nuclear Energy, vol. 154, p. 108113, 2021.
[26] S. Şeker, E. Ayaz, and E. Türkcan, "Elman's recurrent neural network applications to condition monitoring in nuclear power plant and rotating machinery," Engineering applications of artificial intelligence, vol. 16, no. 7-8, pp. 647-656, 2003.
[27] Z. Chen, C. K. Yeo, B. S. Lee, and C. T. Lau, "Autoencoder-based network anomaly detection," in 2018 Wireless Telecommunications Symposium (WTS), 2018: IEEE, pp. 1-5.
[28] L. H. Chiang, E. L. Russell, and R. D. Braatz, Fault detection and diagnosis in industrial systems. Springer Science & Business Media, 2000.
[29] 林廷翰, "核電廠在線智慧監控：肇始事件與失能感測器之偵測與診斷," 博士, 國立清華大學, 2021.
[30] M.-D. Wang, T.-H. Lin, K.-C. Jhan, and S.-C. Wu, "Abnormal event detection, identification and isolation in nuclear power plants using LSTM networks," Progress in Nuclear Energy, vol. 140, p. 103928, 2021.
[31] 王夢蝶, "核電廠未知事件之偵測與辨識," 碩士, 國立清華大學, 2020.
[32] S.-C. Wu, K.-Y. Chen, T.-H. Lin, and H.-P. Chou, "Multivariate algorithms for initiating event detection and identification in nuclear power plants," Annals of Nuclear Energy, vol. 111, pp. 127-135, 2018.
[33] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," Advances in neural information processing systems, vol. 25, 2012.
[34] M. A. Wani, F. A. Bhat, S. Afzal, and A. I. Khan, Advances in deep learning. Springer, 2020.
[35] S. Wang and J. Jiang, "Learning natural language inference with LSTM," arXiv preprint arXiv:1512.08849, 2015.
[36] A. Graves, N. Jaitly, and A.-r. Mohamed, "Hybrid speech recognition with deep bidirectional LSTM," in 2013 IEEE workshop on automatic speech recognition and understanding, 2013: IEEE, pp. 273-278.
[37] R. Caruana, S. Lawrence, and C. Giles, "Overfitting in neural nets: Backpropagation, conjugate gradient, and early stopping," Advances in neural information processing systems, vol. 13, 2000.
[38] M. Verleysen and D. François, "The curse of dimensionality in data mining and time series prediction," in International work-conference on artificial neural networks, 2005: Springer, pp. 758-770.
[39] Y.-G. No, J.-H. Kim, M.-G. Na, D.-H. Lim, and K.-I. Ahn, "Monitoring severe accidents using AI techniques," Nuclear Engineering and Technology, vol. 44, no. 4, pp. 393-404, 2012.
[40] C. S. Burrus, "Introduction to wavelets and wavelet transforms: a primer," Englewood Cliffs, 1997.
[41] H. Abdi and L. J. Williams, "Principal component analysis," Wiley interdisciplinary reviews: computational statistics, vol. 2, no. 4, pp. 433-459, 2010.
[42] T.-H. Lin, S.-C. Wu, K.-Y. Chen, and H.-P. Chou, "Feature extraction and sensor selection for NPP initiating event identification," Annals of Nuclear Energy, vol. 103, pp. 384-392, 2017.
[43] T. Kindred, "Modular Accident Analysis Program 5 (MAAP5) applications guidance: Desktop reference for using MAAP5 softwared — Phase 3 report," Electric Power Research Insitute, 2017.
[44] Y.-H. Cheng, C. Shih, S.-C. Chiang, and T.-L. Weng, "Introducing PCTRAN as an evaluation tool for nuclear power plant emergency responses," Annals of Nuclear Energy, vol. 40, no. 1, pp. 122-129, 2012.
[45] A. J. Spurgin, P. Moieni, and J. P. Spurgin, "Some applications of full scope plant simulators," in Conference Record for 1992 Fifth Conference on Human Factors and Power Plants, 1992: IEEE, pp. 513-519.
[46] AEC, "The Republic of China National Report for the Convention on Nuclear Safety," 2016.