近年來全球太空活動日益頻繁，包括美國、中國、日本及韓國等國皆不斷發射太空載具。過去太空載具發射影響電離層研究顯示太空載具的發射會造成波動，經濾波處理二維全電子含量投射至地圖上可顯示其顯著的效應，根據過去文獻以及本論文在近兩年所觀測的多個事件中可觀測到明顯的電離層擾動，這些擾動在不同事件間雖然有差異但通常展現衝擊波特徵，隨著載具升空軌跡呈現同心圓或V型分布。本研究動機來自於過往研究中我們能確切地在經過濾波後在各GNSS衛星的電離層全電子含量時序圖看到這些衝擊波特徵的擾動，嘗試整理分析這些事件之共同特徵，標記這些擾動的資料點並以監督式機器學習方法使用深度學習演算法訓練，解析模型對於這些擾動特徵的辨識成效。本論文研究使用的深度學習演算法為時間通道卷基網路模型（TCN）和雙向長短期記憶模型（biLSTM），觀察兩個不同邏輯的演算法模型的成效。本研究主要使用資料為西元2021至2023年間東亞與東北亞地區的多起太空活動事件期間的GNSS全電子含量觀測資料，並加入過往研究的西元2010年中國火箭、西元2016年北韓的洲際飛彈事件作為訓練資料。為避免模型誤認將所有電離層震幅劇烈的其他擾動資料視為太空載具所誘發的擾動，也額外增加受電漿泡影響的觀測資料作為訓練資料以增加模型的泛化性。本論文的研究中，我們使用過往研究曾分析發生於北美西岸的西元2016年與西元2017年的火箭發射事件作為驗證模型成效的事件，也將西元2024年1月份於東亞發生共三起火箭事件進行驗證。在這些事件中，我們觀察到TCN模型較biLSTM模型不易誤判其他時段的擾動作為太空載具誘發的擾動，並與參考日的偵測結果差異上TCN模型兩天的偵測差異量較集中於火箭發射後一小時內的時段。我們也進一步比較四種全電子含量資料濾波方法，其中使用高通濾波以及S-G濾波方法的有較低的誤判率。

In recent years, global space activities have become increasingly frequent, with countries like the United States, China, Japan, and North and South Korea continuously launching space vehicles or rockets. Previous studies have shown that launches of the space vehicles significantly affect ionosphere by producing waves. It appears in the filtered two-dimensional total electron content (TEC) maps showing coherent wave structures originated from the rocket trajectory. Additionally, most of events in the past two years exhibit noticeable disturbances, which, although varying between events, generally display shock wave characteristics and are distributed concentrically or in a V-shape along the vehicle's ascent trajectory. Based on these observations of characteristics, we collect multiple space vehicle launce events of East and Northeast Asia during 2021 to 2023, including some of the events mentioned in previous study, such as the 2010 Chinese and 2016 North Korean events, as the training data. These events include satellite launches of China and ballistic missile tests by North Korea. We organized and labeled these data, using supervised machine learning methods to train models to learn various shock wave characteristics, and then evaluated the performance of model using several other events. In this study, we find that the TCN model of deep learning is less prone to falsely identifying the disturbances from other periods as space vehicle-induced disturbances compared to the biLSTM model. Furthermore, the detection discrepancies of TCN model in comparison to the result of reference day are more concentrated within one hour after the rocket launch. Finally, we also compare the impact of the filter methods to the deep learning model performance and among the four filtering methods, the high-pass and S-G filtering methods performed better by giving lower false positive rate for deep learning model.

Afraimovich, E. L., Kosogorov, E. A., & Plotnikov, A. V. (2002). Shock–acoustic waves generated during rocket launches and earthquakes. Cosmic Research, 40, 241-254.
Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.
Bishop, C. M., & Nasrabadi, N. M. (2006). Pattern recognition and machine learning (Vol. 4, No. 4, p. 738). New York: springer.
Choi, B. K., & Kil, H. (2017). Large ionospheric TEC depletion induced by the 2016 North Korea rocket. Advances in Space Research, 59(2), 532-541.
Chou, M. Y., Shen, M. H., Lin, C. C., Yue, J., Chen, C. H., Liu, J. Y., & Lin, J. T. (2018). Gigantic circular shock acoustic waves in the ionosphere triggered by the launch of FORMOSAT‐5 satellite. Space Weather, 16(2), 172-184.
Davies, K., Ionospheric Radio, Short Run Press Ltd., Exeter, England, 1990
Glorot, X., & Bengio, Y. (2010, March). Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intelligence and statistics (pp. 249-256). JMLR Workshop and Conference Proceedings.
Graves, A., Fernández, S., & Schmidhuber, J. (2005, September). Bidirectional LSTM networks for improved phoneme classification and recognition. In International conference on artificial neural networks (pp. 799-804). Berlin, Heidelberg: Springer Berlin Heidelberg.
He, K., Zhang, X., Ren, S., & Sun, J. (2015). Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the IEEE international conference on computer vision (pp. 1026-1034).
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
Hofmann-Wellenhof, B., Lichtenegger, H., & Collins, J. (2012). Global positioning system: theory and practice. Springer Science & Business Media.
Janiesch, C., Zschech, P., & Heinrich, K. (2021). Machine learning and deep learning. Electronic Markets, 31(3), 685-695.
Kelley, M. C. ,(2009). The Earth’s Ionosphere Plasma Physics and Electrodynamics Second Edition.
Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
Lin, C. C., Shen, M. H., Chou, M. Y., Chen, C. H., Yue, J., Chen, P. C., & Matsumura, M. (2017). Concentric traveling ionospheric disturbances triggered by the launch of a SpaceX Falcon 9 rocket. Geophysical Research Letters, 44(15), 7578-7586.
Schafer, R. W. (2011). What is a Savitzky-Golay filter?[lecture notes]. IEEE Signal processing magazine, 28(4), 111-117.
Sarker, I. H. (2021). Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions. SN Computer Science, 2(6), 420.
陳佳宏. (2006). 電離層赤道異常與赤道電噴流. 中央大學太空科學研究所學位論文, 2006, 1-73.
陳威翰. (2013). 利用GPS觀測地震和海嘯對電離層的影響
程品軒. (2020). 研究低緯度之電離層移行擾動與開發自動判別演算法
蔡和芳. (1999). 全球定位系統觀測電離層赤道異常之研究 (Doctoral dissertation, 國立中央大學太空科學研究所博士論文, 桃園).