High-reliable ionospheric prediction information is of great significance for precise positioning and navigation, space weather monitoring, as well as wireless communication (Schunk & Sojka, 1996;Zhang et al., 2022). Many models have been proposed to model the ionosphere and predict its changes. However, the prediction accuracy is limited due to the irregular characteristics and the complicated mechanism of ionospheric variations (Liu et al., 2022;Mcgranaghan et al., 2018). In recent years, the rapidly developed deep learning technique has been steadily applied to ionospheric prediction with an increase in long-term ionospheric observation data (Camporeale, 2019;Ren et al., 2022).Since the mid-1990s, artificial neural networks (ANN), the forerunner of deep learning, provide a new efficient solution for ionospheric modeling and prediction (Joselyn et al., 1993;Oyeyemi et al., 2005). Cander (1998) used different artificial neural networks to simulate and predict the temporal-spatial variations of ionospheric key parameters, such as f0F2 and total electron content (TEC). The prediction results proved the promising prospect of artificial neural networks in ionospheric prediction. Zeng et al. (2002) used Global Positioning System (GPS) occultation data to predict the ionospheric electron density. The results showed that the ANN was an effective

No abstract