簡易檢索 / 詳目顯示

回結果列表
研究生： 黃振庭
Huang, Cheng-Tim
論文名稱： ST-TimNet：結合物理先驗資訊於時空模型之電離層 vTEC 補全與預測
ST-TimNet: Incorporating Physics-Based Priors into Spatio-Temporal Networks for Joint Ionospheric vTEC Imputation and Forecasting
指導教授： 陳奇業
Chen, Chi-Yeh
學位類別： 碩士
Master
系所名稱： 電機資訊學院 - 資訊工程學系
Department of Computer Science and Information Engineering
論文出版年： 2026
畢業學年度： 114
語文別： 英文
論文頁數： 60
中文關鍵詞： 電離層垂直總電子含量 、全球導航衛星系統 、資料補全 、時空預測 、物理引導深度學習 、國際參考電離層 、空間錨點 、隨機損失權重
外文關鍵詞： ionospheric vTEC, GNSS, missing data imputation, spatiotemporal forecasting, physics-guided deep learning, International Reference Ionosphere, spatial anchor, Random Loss Weighting
相關次數： 點閱：4  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 電離層垂直總電子含量（vTEC）是影響全球導航衛星系統（GNSS）定位精度的關鍵。然而，實際觀測資料受海洋限制與衛星穿刺點持續移動影響，常呈現嚴重的時空缺失，在台灣中低緯度區域更具快速且非線性的動態變化。既有研究多將歷史缺失補全與未來預測視為獨立任務，易導致誤差傳遞且難以整合時空結構。為此，本研究提出ST-TimNet，一個結合物理先驗與深度時空網路的端到端統一框架，將稀疏歷史觀測、國際參考電離層（IRI）物理先驗及多尺度傅立葉時間特徵整合於單一架構。模型首先透過補全模型重建歷史資料並產生具時空語意的特徵，預測模型再結合未來 IRI 與時間特徵在選定的空間錨點上進行預測。為降低長時間預測的計算成本，本框架採用空間錨點與補全機制，將部份未來預測藉由共享的補全模型還原為完整網格資料，首次在單一網路中實現兩項任務的學習。
    本研究以台灣及周邊區域（北緯 15°–30°、東經 115°–130°）進行實驗，網格解析度為 0.5° × 0.5°，並用五分鐘為間隔，採樣經克里金方法內插處理後的觀測數據作為訓練與評量的輸入與輸出。模型目標是在五分鐘時間解析度下，補全整日二十四小時的缺失資料，並以重建後的資料預測隔日二十四小時的電離層垂直總電子含量，同時引入 IRI-2020 電離層垂直總電子含量作為先驗資訊。模型透過兩階段訓練策略與隨機損失權重平衡雙任務損失，並結合頻域正則化提升長時間預測穩定性。實驗結果顯示，ST-TimNet 在電離層垂直總電子含量缺失資料補全上達到均方根誤差 RMSE = 1.95 TECU，並在二十四小時預測任務上達到 RMSE = 11.82 TECU。整體而言，ST-TimNet 成功在高時空解析度下同時完成電離層垂直總電子含量補全與整日預測，且表現優於一些模型。此框架為台灣區域提供了適用的高解析度時空建模框架，能有效支援近即時的全球導航衛星系統誤差修正與太空天氣監測需求。

    Ionospheric vertical Total Electron Content (vTEC) is critical for Global Navigation Satellite System (GNSS) positioning accuracy. However, real-world observations suffer from severe spatiotemporal missingness due to ocean gaps and continuous satellite movement, exhibiting rapid and nonlinear dynamics especially over the low-to-middle latitude region of Taiwan. Existing studies typically treat historical data imputation and future forecasting as separate tasks, which propagates errors and fails to exploit shared spatiotemporal structures. To address these limitations, this thesis proposes ST-TimNet, a unified end-to-end framework that integrates physics-guided modeling with deep learning by incorporating sparse observations, International Reference Ionosphere (IRI) priors, and multi-scale Fourier time embeddings within a single architecture. The imputation module reconstructs incomplete historical fields into latent spatiotemporal tokens, which the forecast module uses alongside future IRI and temporal features to predict future vTEC on selected spatial anchor nodes. To reduce computational costs, ST-TimNet adopts a spatial-anchor and imputer-completion strategy where sparse future anchor predictions are scattered back and completed by reusing the shared imputation module, enabling joint learning of both tasks under a single network.
    Experiments are conducted over Taiwan and its surrounding region (15°N–30°N, 115°E–130°E) discretized into a 0.5° × 0.5° regular grid, sampling observation data processed by Kriging interpolation (Kriging vTEC) at a 5-minute interval for training and evaluation inputs and outputs. The model aims to impute a full-day 24-hour sequence of missing observations and use this reconstructed data to forecast the subsequent 24-hour vTEC, both at a 5-minute temporal resolution, while utilizing IRI-2020 vTEC as the exogenous physics prior. The model is optimized via a two-stage training strategy with Random Loss Weighting (RLW) and frequency-domain regularization to improve long-horizon stability. Experimental results demonstrate that ST-TimNet achieves an imputation RMSE of 1.95~TECU and a 24-hour forecasting RMSE of 11.82~TECU. Overall, ST-TimNet successfully performs vTEC imputation and full-day forecasting at high spatial and temporal resolutions, outperforming some models. ST-TimNet provides a high-resolution spatiotemporal modeling framework suitable for regional GNSS error correction and near-real-time space weather monitoring over Taiwan.

    中文摘要 i Abstract iii 誌謝 v Contents vi List of Tables viii List of Figures ix 1 Introduction 1 1.1 Background and Motivation 1 1.2 Research Challenges and Problem Statement 2 1.3 Contributions 3 2 Related Work 5 2.1 Classical Interpolation for Ionospheric vTEC 5 2.2 Heuristic and Early Neural Network Approaches 6 2.2.1 Early Machine Learning Models for Enhancing Numerical Methods 6 2.2.2 The Era of Deep Learning Models to Capture Timeseries Change 6 2.2.3 Fusion of Spatial and Temporal Features 7 2.3 Advancements in Spatiotemporal Models 7 2.3.1 Deep Learning for Data Imputation and Gap Completion 8 2.3.2 Advanced Architectures for Long-Horizon Forecasting 8 2.4 Identified Research Gaps and Positioning 9 3 Method 10 3.1 Overview of ST-TimNet 10 3.1.1 Problem Setting 10 3.1.2 Model Architecture 11 3.2 Data Preparation and Feature Engineering 12 3.3 Imputation Module 15 3.3.1 Temporal-Spatial Interaction 17 3.4 Forecast Module 18 3.4.1 TimeXer Encoder 19 3.4.2 Sparse-Anchor Future Completion 19 3.5 Training Objectives 20 3.5.1 Imputation Loss 21 3.5.2 Forecast Loss 21 3.5.3 Random Loss Weighting and Joint Optimization 23 4 Experiments 25 4.1 Dataset 25 4.2 Evaluation Metrics 28 4.3 Implementation Details 29 4.4 Main Results 31 4.4.1 Missing Data Imputation Performance 31 4.4.2 24-Hour Forecasting Performance 33 4.5 Robustness and Case Studies 36 4.5.1 Robustness to Observation Sparsity 36 4.5.2 Representative Imputation and Forecast Case Studies 37 4.5.3 Hourly and Horizon-wise Error Analysis 38 4.6 Ablation Study 39 4.6.1 Effect of Physical Prior and Temporal Context 40 4.6.2 Effect of Joint Training and Loss Balancing 41 5 Conclusions 43 References 45

    [1] L. Liu, S. Zou, Y. Yao, and E. Aa, “Multi-scale ionosphere responses to the may 2017 magnetic storm over the asian sector,” GPS Solutions, vol. 24, no. 1, p. 26, 2020.
    [2] A. Komjathy, Global ionospheric total electron content mapping using the Global Positioning System. University of New Brunswick Fredericton, 1997, vol. 188.
    [3] S. Dubey, R. Wahi, and A. Gwal, “Ionospheric effects on gps positioning,” Advances in Space Research, vol. 38, no. 11, pp. 2478–2484, 2006.
    [4] A. Rovira-Garcia, D. Ibáñez-Segura, R. Orús-Perez, J. M. Juan, J. Sanz, and G. González-Casado, “Assessing the quality of ionospheric models through gnss positioning error: Methodology and results,” GPS solutions, vol. 24, no. 1, p. 4, 2020.
    [5] I. Sayin, F. Arikan, and O. Arikan, “Regional tec mapping with random field priors and kriging,” Radio Science, vol. 43, no. 05, pp. 1–14, 2008.
    [6] D. Yang, H. Fang, and Z. Liu, “Completion of global ionospheric tec maps using a deep learning approach,” Journal of Geophysical Research: Space Physics, vol. 127, no. 5, e2022JA030326, 2022.
    [7] L. Liu, S. Zou, Y. Yao, and Z. Wang, “Forecasting global ionospheric tec using deep learning approach,” Space Weather, vol. 18, no. 11, e2020SW002501, 2020.
    [8] T. Nie, G. Qin, W. Ma, Y. Mei, and J. Sun, “Imputeformer: Low rankness-induced transformers for generalizable spatiotemporal imputation,” Proceedings of the 30th ACM SIGKDD conference on knowledge discovery and data mining, 2024, pp. 2260–2271.
    [9] J. Liu, W. Yang, C. Lin, Y. Chen, and I. Lee, “A statistical study on the characteristics of ionospheric storms in the equatorial ionization anomaly region: Gps-tec observed over taiwan,” Journal of Geophysical Research: Space Physics, vol. 118, no. 6, pp. 3856–3865, 2013.
    [10] T. Tsugawa, M. Nishioka, M. Ishii, K. Hozumi, S. Saito, A. Shinbori, Y. Otsuka, A. Saito, S. M. Buhari, M. Abdullah, et al., “Total electron content observations by dense regional and worldwide international networks of gnss,” Journal of Disaster Research, vol. 13, no. 3, pp. 535–545, 2018.
    [11] P. Rama Rao, K. Niranjan, D. Prasad, S. Gopi Krishna, and G. Uma, “On the validity of the ionospheric pierce point (ipp) altitude of 350 km in the indian equatorial and low-latitude sector,” Annales Geophysicae, Copernicus Publications Göttingen, Germany, vol. 24, 2006, pp. 2159–2168.
    [12] K. Davies and G. Hartmann, “Studying the ionosphere with the global positioning system,” Radio Science, vol. 32, no. 4, pp. 1695–1703, 1997.
    [13] Y. Wang, H. Wu, J. Dong, G. Qin, H. Zhang, Y. Liu, Y. Qiu, J. Wang, and M. Long, “Timexer: Empowering transformers for time series forecasting with exogenous variables,” Advances in Neural Information Processing Systems, vol. 37, pp. 469–498, 2024.
    [14] B. Lin, F. Ye, Y. Zhang, and I. Tsang, “Reasonable effectiveness of random weighting: A litmus test for multi-task learning,” Transactions on Machine Learning Research.
    [15] D. Bilitza, M. Pezzopane, V. Truhlik, D. Altadill, B. W. Reinisch, and A. Pignalberi, “The international reference ionosphere model: A review and description of an ionospheric benchmark,” Reviews of geophysics, vol. 60, no. 4, e2022RG000792, 2022.
    [16] M. P. Foster and A. N. Evans, “An evaluation of interpolation techniques for reconstructing ionospheric tec maps,” IEEE Transactions on Geoscience and Remote Sensing, vol. 46, no. 7, pp. 2153–2164, 2008.
    [17] M. Li, Y. Yuan, N. Wang, Z. Li, X. Liu, and X. Zhang, “Statistical comparison of various interpolation algorithms for reconstructing regional grid ionospheric maps over china,” Journal of Atmospheric and Solar-Terrestrial Physics, vol. 172, pp. 129–137, 2018.
    [18] M. Rodríguez-Bouza, C. Paparini, X. Otero, M. Herraiz, S. M. Radicella, O. E. Abe, and G. Rodríguez-Caderot, “Southern european ionospheric tec maps based on kriging technique to monitor ionosphere behavior,” Advances in Space Research, vol. 60, no. 8, pp. 1606–1616, 2017.
    [19] R. Zhang, H. Li, Y. Shen, J. Yang, W. Li, D. Zhao, and A. Hu, “Deep learning applications in ionospheric modeling: Progress, challenges, and opportunities,” Remote Sensing, vol. 17, no. 1, p. 124, 2025.
    [20] J. Habarulema and L.-A. McKinnell, “Investigating the performance of neural network backpropagation algorithms for tec estimations using south african gps data,” Annales geophysicae, Copernicus Publications Göttingen, Germany, vol. 30, 2012, pp. 857–866.
    [21] Z. Huang, Q. Li, and H. Yuan, “Forecasting of ionospheric vertical tec 1-h ahead using a genetic algorithm and neural network,” Advances in Space Research, vol. 55, no. 7, pp. 1775–1783, 2015.
    [22] R. Song, X. Zhang, C. Zhou, J. Liu, and J. He, “Predicting tec in china based on the neural networks optimized by genetic algorithm,” Advances in Space Research, vol. 62, no. 4, pp. 745–759, 2018.
    [23] I. Srivani, G. S. V. Prasad, and D. V. Ratnam, “A deep learning-based approach to forecast ionospheric delays for gps signals,” IEEE Geoscience and Remote Sensing Letters, vol. 16, no. 8, pp. 1180–1184, 2019.
    [24] W. Sun, L. Xu, X. Huang, W. Zhang, T. Yuan, Z. Chen, and Y. Yan, “Forecasting of ionospheric vertical total electron content (tec) using lstm networks,” 2017 international conference on machine learning and cybernetics (ICMLC), IEEE, vol. 2, 2017, pp. 340–344.
    [25] L. Liu, Y. J. Morton, and Y. Liu, “Ml prediction of global ionospheric tec maps,” Space Weather, vol. 20, no. 9, e2022SW003135, 2022.
    [26] A. Ruwali, A. S. Kumar, K. B. Prakash, G. Sivavaraprasad, and D. V. Ratnam, “Implementation of hybrid deep learning model (lstm-cnn) for ionospheric tec forecasting using gps data,” IEEE Geoscience and Remote Sensing Letters, vol. 18, no. 6, pp. 1004–1008, 2020.
    [27] S.-H. Jeong, W. K. Lee, S. Jang, H. Kil, J.-H. Kim, Y.-S. Kwak, Y. H. Kim, J. Hong, and B.-K. Choi, “Reconstruction of the regional total electron content maps over the korean peninsula using deep convolutional generative adversarial network and poisson blending,” Space Weather, vol. 20, no. 8, e2022SW003131, 2022.
    [28] X. Lin, H. Wang, Q. Zhang, C. Yao, C. Chen, L. Cheng, and Z. Li, “A spatiotemporal network model for global ionospheric tec forecasting,” Remote Sensing, vol. 14, no. 7, p. 1717, 2022.
    [29] S.-H. Jeong, W. K. Lee, H. Kil, S. Jang, J.-H. Kim, and Y.-S. Kwak, “Deep learning-based regional ionospheric total electron content prediction—long short-term memory (lstm) and convolutional lstm approach,” Space Weather, vol. 22, no. 1, e2023SW003763, 2024.
    [30] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” vol. 30, 2017.
    [31] S. Liu, E. Johns, and A. J. Davison, “End-to-end multi-task learning with attention,” Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 1871–1880.
    [32] W. Du, D. Côté, and Y. Liu, “Saits: Self-attention-based imputation for time series,” Expert Systems with Applications, vol. 219, p. 119619, 2023.
    [33] A. Das, W. Kong, A. Leach, S. K. Mathur, R. Sen, and R. Yu, “Long-term forecasting with tide: Time-series dense encoder,” Transactions on Machine Learning Research.
    [34] Y. Liu, T. Hu, H. Zhang, H. Wu, S. Wang, L. Ma, and M. Long, “Itransformer: Inverted transformers are effective for time series forecasting,” International conference on learning representations, vol. 2024, 2024, pp. 11116–11140.
    [35] Y. Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, “A time series is worth 64 words: Long-term forecasting with transformers,” The Eleventh International Conference on Learning Representations.
    [36] S. Wang, H. Wu, X. Shi, T. Hu, H. Luo, L. Ma, J. Zhang, and J. Zhou, “Timemixer: Decomposable multiscale mixing for time series forecasting,” International conference on learning representations, vol. 2024, 2024, pp. 38626–38652.
    [37] K. Sivakrishna, D. V. Ratnam, and G. Sivavaraprasad, “A bidirectional deep-learning algorithm to forecast regional ionospheric tec maps,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 4531–4543, 2022.
    [38] P. Galav, N. Dashora, S. Sharma, and R. Pandey, “Characterization of low latitude gps-tec during very low solar activity phase,” Journal of Atmospheric and Solar-Terrestrial Physics, vol. 72, no. 17, pp. 1309–1317, 2010.
    [39] Z. Chen, W. Liao, H. Li, J. Wang, X. Deng, and S. Hong, “Prediction of global ionospheric tec based on deep learning,” Space Weather, vol. 20, no. 4, e2021SW002854, 2022.

    QR CODE