1. Introduction

In recent years, the rising need for accurate space weather forecasting has gained significant attention, especially in understanding and countering the disruptive effects of solar activity on Earth’s ionosphere. Total electron content (TEC) is a key metric that quantifies electron density available in the ionosphere and is vital for satellite communication and navigation systems. With the increasing occurrence of various classes of solar flares (SFs) ejected from the sun, ionospheric TEC anomalies will occur. Among these, X-class SFs, characterized by a peak intensity of I ≥ 10⁻⁴ watts per square meter, are the most powerful and can cause severe disturbances to both satellite operations and ionospheric electron content. The impact of SF on the ionosphere leads to an increase in TEC. Therefore, the accurate prediction of TEC fluctuations will be useful to mitigate the errors in satellite communication. Several AI models used for TEC prediction are deliberated in various kinds of literature studies. Some of the important literature studies are discussed in this paper. Yasyukevich et al. [1] studied the ionospheric response during two SFs (X2.2 and X9.3) that occurred on September 6, 2017, using data obtained from GNSS satellites and HF radio communication systems. They analyzed TEC variations based on HF propagation and observational data. The X2.2 flare caused a 3–4 TECU increase at midlatitudes, whereas the X9.3 flare caused a larger 8–16 TECU increase at low latitudes. The flares also caused HF blackouts, but no significant GNSS signal loss, though GPS positioning errors increased threefold during the X9.3 event. Sezen et al. [2] focused on predicting near-real-time TEC using the IONOLAB-TEC system, which provides TEC estimates from GPS data collected from IGS and EUREF stations. The data are processed using the Reg-Est algorithm through IONOLAB. Users can compute TEC for one or multiple stations and compare values across different days. The model predicts TEC accurately by accounting for biases and mapping functions. The system is reliable and user-friendly for monitoring space weather events. Ren et al. [3] used a mixed CNN–Bi-LSTM model to forecast the ionospheric TEC during ionospheric storms. They collected 25 years of data from 1998 to 2023, specifically using global ionosphere map (GIM)-TEC along with geomagnetic indices such as Kp, Ap, and Dst. The mixed model predicted well when compared to the Bi-LSTM-DNN model, especially in short-term forecasting. It captured the process of ionospheric storms but showed some prediction errors when extending the forecast length. The accuracy was higher in the solar–geographical reference frame and daytime predictions. Guyer and Can [4] predicted TEC changes caused by M- and X-class SFs based on the data collected during the year 2012, particularly in July. They used dual-frequency GPS satellite systems to analyze ionospheric disturbances before, during, and after significant flares. Key flares examined included an M5.6 flare on July 2 and an M6.1 flare on July 5. The study focused on both middle and high latitudes, particularly using the Mizu IGS station in Japan for analysis. Results indicated a negative relationship between TEC changes and the solar zenith angle, with significant TEC enhancements during flare events.

Zhou et al. [5] investigated the effect of SFs and geomagnetic storms on GNSS performance based on the collected data from September 2017 to October 2021. The data include solar activity parameters, TEC index maps, and GNSS observations from 39 IGS stations. They analyzed the SiS ranging error and detected cycle slips in GNSS phase observations. The analysis revealed that cycle slips increased during geomagnetic storms, leading to degraded positioning accuracy, particularly in low- and high-latitude areas. Wu et al. [6] used the data taken from IGS GIMs and the Lowell GIRO Data Center to study the impact of SFs and geomagnetic storms on the ionosphere in Western Europe during March 23–26, 2024. They found significant ionospheric anomalies, especially in high-latitude regions, caused by geomagnetic storms rather than SFs alone. The study highlighted disruptions such as decreased TEC and uneven electron density during the storm. Senturk [7] analyzed the global ionospheric response to the June 2015 geomagnetic storm using TEC data obtained from CODE GIMs, COSMIC radio occultation (RO), and IGS stations. The author examined both positive and negative TEC phases across different latitudes, highlighting longitudinal and hemispherical asymmetries. The analysis showed intense positive phases in the Southern Hemisphere (SH) and negative phases in the Northern Hemisphere (NH). COSMIC RO data revealed that the ionospheric response depends not only on TEC variations but also on the altitude of maximum electron density.

Sivavaraprasad et al. [8] developed a forecasting algorithm for the prediction of ionospheric TEC irregularities in Bengaluru, India, using the data taken from the IISC station from 2009 to 2016. They combined PCA with linear and autoregressive moving average (ARMA) and neural network (NN) models, finding that the NN model had better accuracy during geomagnetic quiet periods. The results indicated an MAE of 0.5 TECU/hour in quiet conditions and 0.98 TECU/hour during disturbed conditions. Singh et al. [9] predicted the impact of powerful SFs on TEC variability at a low-latitude station in Varanasi, India, focusing on flare events from 2015 to 2017. They used GPS observations to collect data, particularly during SFs in March 2015, January and February 2016, August 2016, and September 2017. No specific predictive model is mentioned, but they analyzed TEC responses considering geomagnetic disturbances and solar zenith angles. The results showed TEC enhancements of up to 15 TECU, with stronger effects during midday hours. Hazarika et al. [10] studied the impact of 11 X-class SFs (2009–2013) from solar cycle 24 on the ionosphere at Dibrugarh, India, using TEC data derived from GPS signals. They detected a nonlinear correlation (R² = 0.86) among EUV and TEC augmentations, driven by similar nonlinearity between X-ray and EUV fluxes. The local time of the flares influenced TEC magnitudes while correcting X-ray flux for zenith angle effects did not improve the correlation. Singh et al. [11] studied VLF wave amplitude measurements at Varanasi, India (2011–2012), to analyze SF impacts on the D-region ionosphere during the increasing phase of solar cycle 24. Using data from the NWC transmitter and GOES satellite, they found that SFs significantly lowered the ionospheric reflection height (H) and increased the sharpness factor (β). The results showed more pronounced electron density changes during X-class flares, with different dynamics compared with midlatitudes.

Barta et al. [12] investigated SF impacts on ionospheric absorption using ionograms from mid- and low-latitude stations during eight X- and M-class flares that happened in solar cycle 23. They used data from ionosonde stations across Europe and South Africa (2001–2006) and applied the dfmin parameter as a measure of nondeviation absorption. Their results showed that radio fade-outs and absorption were strongly dependent on solar zenith angle, with larger absorption at smaller zenith angles. The D-RAP model was used as a reference to explain these variations. Wu et al. [6] investigated the impact of SF and geomagnetic storms on the ionosphere, focusing on February, March, May, June, July, and August 2024 events. Using data from the Lowell GIRO Data Center and IGS GIMs, they found significant geomagnetic storms following the SF on March 23, 2024. These storms caused widespread reductions in TEC, foF2, and m3000F2 values in Western Europe. High-latitude regions experienced more severe effects, whereas low-latitude areas such as Spain showed increased ionospheric jitter. The results emphasized that geomagnetic storms are crucial drivers of ionospheric anomalies, alongside solar activity. Saharan et al. [13] investigated the ionospheric response to 450 SFs (5 X-class SFs, 49 M-class SFs, and 396 C-class SFs) that occurred in 2014, which coincided with the maximum phase of solar cycle 24, using GPS-TEC data from the IGS station in Bangalore, India. They examined the correlation between differential vertical TEC (DDVTEC) and X-ray/EUV fluxes for X-, M-, and C-class flares, employing both baseline and mean methods. The baseline method revealed a stronger correlation for X-class flares, showing a correlation of 0.606 for X-ray and 0.67 for EUV flux. In contrast, no significant correlation was found for M- and C-class flares. The study indicates that the results were influenced by high solar activity and the specific site of the station. Davoudifar et al. [14] analyzed the TEC changes during the 24th solar cycle based on the GPS-TEC data obtained from the Tehran, midlatitude station. A semiempirical model was developed to obtain the mean values, and those values were compared with the IRI model. The results show that during X-class SFs, 20% variations in TEC were noticed. Reddybattula et al. [15] used LSTM for TEC prediction based on 8 years of training data from 2009 to 2017. The predicted results of the LSTM were compared with the IRI-2016. The outcomes during the test period year (2018) LSTM closely followed the actual GPS-TEC data with a minimal root-mean-square error (RMSE) of 1.6149 and the highest CC of 0.992. Tang et al. [16] used ARMA, LSTM, and Seq2Seq models to forecast the TEC during different geomagnetic storm periods based on the MIT madrigal observation from 2001 to 2016. Among the three models, LSTM performed well during strong geomagnetic storms. Kiruthiga et al. [17] forecasted TEC using OKSM over the low-, mid-, and high-latitude regions during the X9.3 SF that occurred in September 2017. The results were compared with the IRI-2016 and PLAS 2017 models. The results show that during the extreme solar disturbance, the OKSM predicted well. Mukesh et al. [18] performed TEC forecasts for three latitude stations using the COK model during storms and for the IISC station during high solar activity days. The predicted results were compared with the IRI-Plas 2017 and EOF models. The results reveal that the COK model gives good predictions when compared to other models. Alemu et al. [19] used SVM and LSTM for the prediction of TEC over Ethiopia during quiet and storm geomagnetic conditions based on the TEC data taken from the ADIS, CURG, and NEGE stations during the years 2013 to 2016. The predicted results were compared with the IRI-2016. The results specified that the SVM model predicted better than the LSTM and IRI models.

Long [20] introduces an integrated model for an energy-efficient building envelope plan. The model employed ML algorithms such as RF, ANN, DNN, SVM, GENLIN, and GB to evaluate energy consumption. The AI optimization algorithms such as NSGA II, DSE, and MOPSO were integrated with the ML algorithms during the evolutionary process for obtaining optimal predictive results. Arunachalam et al. [21] predicted the QoS data using the DeepAR algorithm combined with 11 optimizers based on the data obtained from various sensors. The results indicate that both the FTML and RMSprop optimizers performed well in optimizing the weights of the DeepAR to obtain better results. Weng et al. [22] combined the MMAdapGA optimizer with the BP-NN algorithm for the prediction of TEC based on the data obtained from the Athens station in Greece. The predicted results were compared with the IRI model and other AI models. The comparative results show that the optimizer-combined AI model provides higher accuracy. After a detailed study of the various literature studies, this paper attempts to integrate the optimizer with the AI model to predict the TEC during the X-class SFs that occurred in the year 2024.

This research explores the prediction of TEC during notable X-class SFs that occurred in May 2024 using the Bi-LSTM coupled with the Adam optimizer (Bi-LSTM-AO) [23]. The investigation focuses on several significant X-class SFs, including an X8.79 flare on May 14, an X6.3 flare on February 22, an X4.52 flare on May 6, an X3.98 flare on May 10, an X2.99 flare on May 27, an X2.0 flare on July 16, an X1.72 flare on May 14, an X1.7 flare on August 5, an X1.6 flare on May 3, an X1.43 flare on June 1, an X1.2 flare on May 14, and an X1.12 flare on March 28, 2024. Each of these powerful solar events was analyzed to assess the performance of the Bi-LSTM model in forecasting TEC variations under extreme space weather conditions. The graphical representation of the effects of X-class SFs on the ionosphere and GPS signals is illustrated in Figure 1, providing a visual context to the analysis conducted in this research.