Machine learning-based GPS TEC forecasting

Leow, Hock Yew (2024) Machine learning-based GPS TEC forecasting. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full text) Machine learning-based GPS TEC forecasting.pdf - Submitted Version Download (1MB)

Abstract

Precise ionosphere total electron content (TEC) is crucial for various applications such as space weather prediction, satellite communication and navigation, ionosphere scintillation monitoring, and magnetic storm monitoring and so on. Forecasting ionosphere TEC is crucial because by predicting the short-term value of the TEC, it is capable in filling the empty gap between the TEC product latency and increase the precision of the TEC values. In this project, a single type of machine learning model is focused on which is the feed-forward backprop artificial neural network (ANN) and the Levenberg-Marquardt algorithm is used as the training technique. The machine is trained using 12 months of GPS data from FKEKK station together with 12 months of sun activity and magnetic activity from NASA which are the sunspot no, proton ratio, F10.7 index, Kp index, Dst Index, and Ap Index. The MATLAB software is used to train the machine learning algorithm. Net1 which is the neural network model which comprises of all the mentioned solar and geomagnetic input shows the best accuracy with Mean Absolute Error (MAE) of 2.35 TECU, Root Mean Square Error (RMSE) of 3 and R-square of 0.966 which is more accurate than the rest of the neural network models (Net2, Net3, and Net2016). This project applies that Net1 is capable in accurately forecasting the TEC in the southeast Asia region.

Actions (login required)

View Item