The ionospheric total electron content (TEC) plays an important role in the analysis of ionospheric effects and has significant implications for mitigating positioning errors in global navigation satellite systems (GNSS). This paper proposes an innovative neural network model, TECFormer, which integrates the auto-correlation and self-attention mechanisms to predict ionospheric TEC with higher accuracy and stability. TECFormer comprises three components: Trend Prediction, Relation Analysis, and Series Merging. The Trend Prediction leverages the auto-correlation mechanism to roughly predict the trends of TEC and space environment parameters in a series-wise way. Simultaneously, the Relation Analysis utilizes the self-attention mechanism to extract the interaction among the TEC, space environment parameters and geographic location. In the Series Merging, the coarse TEC trend is refined using interrelation information derived from the Relation Analysis to generate more accurate TEC series. In order to verify the effectiveness of TECFormer, four metrics—root mean square error, mean absolute error, mean relative error, and Pearson correlation coefficient—were selected to evaluate its prediction performance and compare the results with other TEC prediction models such as the Autoformer model, the LSTM model, the LAM-LSTM model, the global ionospheric TEC map (GIM), and the Klobuchar model. The experimental results indicate that TECFormer exhibits superior predictive accuracy and robust prediction performance, offering a valuable tool for enhancing the reliability of GNSS.
YU et al. (Mon,) studied this question.