Study On Time Series Prediction Of Ionospheric Total Electron Content Based On Deep Learning

With the advancement of communication technology and the advent of the information age,communication and navigation have become an indispensable part of people's lives.During intense geomagnetic storms,the structure of the ionosphere usually changes dramatically,which will seriously affect radio communication and Satellite navigation system.Ionospheric TEC(Total Electron Content)can well reflect the behavior and state of the ionosphere.In order to monitor the disturbance and adverse effects of the ionosphere under extreme space weather conditions,it is critically important to make accurate ionospheric TEC prediction.However,the prediction of the ionosphere during storms has always been a challenge because the behavior of the ionosphere varies greatly with different magnetic storms,and purely physical methods often fail to obtain satisfactory results.With the continuous improvement of deep learning algorithms,deep learning models have been widely used in temporal prediction research.Therefore,in order to compare the prediction effect of the deep learning model and other models on the ionosphere TEC,and find an effective prediction of the ionosphere in storm TEC method,the main research work of this paper is as follows.We select plenty of geomagnetic storm events based on the MIT madrigal observation from 2001 to 2016 in this article,and study to use a traditional statistical time series prediction method(SARIMA)and two deep learning algorithms(LSTM and Seq2Seq)to predict the storm-time ionospheric TEC under different conditions of geomagnetic storms.And we analyze the statistical error comparison and correlation of multiple events among the three methods.The statistical analysis shows that the LSTM can achieve the best prediction accuracy(more than 70%accuracy)and is robust for the accurate trend prediction of the strong geomagnetic storms.In contrast,SARIMA and Seq2Seq have relatively poor performance with prediction accuracy for the strong geomagnetic storms of 58.18%and 45.45%,respectively.In addition,in order to monitor the ionospheric behavior in more areas,we have proposed related extended research and discussions,using the LSTM model and the LSTM-DCGAN hybrid model to make short-term predictions of the global TEC map,and comparing their respective prediction results with the real TEC The map is compared,and it is found that the prediction accuracy using the LSTM-DCGAN hybrid model is higher than that of the LSTM model.