Temporal Modeling and Prediction of Global Ionospheric Total Electron Content: Applications for Attention Mechanisms
Xingyu Liu, School of Computer Science, Beihang University; Yufei Wang, School of Computer Science, Beihang University;Kunlin Yang, School of Instrumentation and Opto-electronic Engineering, Beihang University; Yang Liu, School of Instrumentation and Opto-electronic Engineering, Beihang University, Marconi Lab, Science Technology and Innovation Section Abdus Salam International Centre for Theoretical Physics
Location: Beacon B
The global Total Electron Content (TEC) is a critical parameter in ionospheric research, as it significantly affects the propagation of trans-ionosphere radio waves, consequently, the performance of satellite communication and navigation systems are also influenced by status of ionosphere. The variations in ionosphere can cause signal delays and attenuation. To address the problem, it is necessary to make accurate prediction of ionosphere TEC so as to strengthen the spatial communication stability and GNSS based navigation precision. Additionally, large fluctuations in ionosphere are closely associated with solar and geomagnetic activities, such as solar and geomagnetic storms, leading to some difficulties for the accurate modeling and forecasting in ionosphere TEC forecasting, especially in the equatorial ionization anomaly region.
With the rapid development of deep learning techniques, time-series prediction models based on deep learning have shown great potential in TEC forecasting. Compared to traditional methods, deep learning models have many advantages in capturing complex spatiotemporal dependencies in TEC series. To address the problem, this work compares the performance of several deep learning models in application of ionosphere TEC prediction, focusing particularly on the usage of attention mechanisms on prediction accuracy. With several representative deep learning models constructed and evaluated, the potential of attention mechanism in TEC series modeling and prediction are well studied and discussed. In the experiments, a number of leading deep learning models are evaluated and tested for short-term ionosphere TEC prediction, including the long-short term memory (LSTM) as well as those attention-based networks like Transformer, TimesNet, N-BEATS, and N-HITS. Two years of TEC series are selected for testing, the year 2014 as the solar maximum in solar cycle 24, and the year 2017 in the solar descending phase of solar cycle 24. The advantage of attention-mechanism are analyzed and compared.
The results show that the Transformer model has significant improvement in prediction accuracy, effectively capturing spatiotemporal variations in TEC. In contrast, TimesNet excels in computational speed, making it ideal for real-time applications where efficiency is paramount. Models with attention mechanisms outstands in prediction accuracy and model reliability. By dynamically adjusting the hyper-parameters in different parts of the input sequence, the attention mechanism helps to focus more on the time points and features that have a greater impact on the prediction results, thus effectively improving the prediction performance.
For Attendees Technical Program Tutorials Registration Hotel Travel and Visas Exhibit Hall For Authors Abstract Management Editorial Review Policies Publication Ethics Policies Author Resource Center For Exhibitors Exhibitor Resource Center Marketing Toolkit Other Years Future Meetings Past Meetings