Forecasting total electron content (TEC) using CEEMDAN LSTM model

The forecasting of ionospheric Total Electron Content (TEC) is necessary for initiating measures to improve the performance of GNSS systems in modern technological infrastructures and applications. The TEC signal derived from GNSS signals is nonstationary and nonlinear due to temporal and spatial variations. This study presents a hybrid CEEMDAN LSTM model for predicting the TEC signals. The CEEMDAN technique reduces the non-linearity of the TEC signal by decomposing it into several intrinsic mode functions (IMFs) which are then predicted by the LSTM network with better accuracy. The proposed model was tested on the GNSS data from the IGS LHAZ, POL2 & STK2 stations to forecast TEC. The prediction results of the proposed model were compared with Neural Network, LSTM, and the International Reference Ionosphere (IRI) model. The RMSE and MAE of the predictions from the proposed model were observed to be 50% and 70% better compared to LSTM and the Neural Network models.