LSTM Based Regional Ionosphere Prediction Modeling Of Total Electronic Content

With the increasing frequency of human activities in outer space,the study of the space atmosphere has received more and more attention.The ionosphere,as an important part of the space atmosphere,has gradually become a research focus of scholars.The satellite navigation GNSS system is currently the most reliable tool for ionospheric detection and research.Using its observation data,especially GPS-based data,a prediction model for ionospheric total electron content can be established to improve ionospheric delay and ionospheric space weather.Forecasts provide important support.This article focuses on the acquisition,analysis,modeling,and prediction of ionospheric total electron content VTEC data.It focuses on the modeling and prediction of VTEC values,and uses the VTEC value predicted by the CODE data analysis center of IGS as a one-day prediction spherical harmonic model as Basic forecasting model.Due to the large error of the CODE spherical harmonic model for the VTEC value in the Japanese region,the historical ionospheric VTEC data of four grids near the Japanese region was used to construct the regional ionospheric prediction based on the linear time series ARMA model and the BP neural network model.The model has a certain improvement in the prediction accuracy.The ARMA model does not consider the nonlinear structure of VTEC data and the BP network model can not fully exploit the shortcomings of VTEC data history information.A long-term and short-term memory LSTM model based on recurrent neural network RNN structure is introduced.The LSTM model was evaluated using a five-day test set.The estimated RMSE of the LSTM model was 0.8 TECU.The average absolute error MAE was 0.59 TECU and the average relative error percentage MAPE was 8.83%.Through experimental comparison and analysis,under the same data set,the performance accuracy of LSTM model relative to CODE spherical harmonic model,ARMA model,and BP neural network model on RMSE,MAE,and MAPE has been improved to some extent.