Long Short?Term Memory Neural Network for Ionospheric Total Electron Content Forecasting Over China

An increasing number of terrestrial- and space-based radio-communication systems are influenced by the ionospheric space weather, making state increasingly important to forecast. In this study, a novel extended encoder-decoder long short-term memory (ED-LSTME) neural network, which can predict total electron content (TEC) is proposed. Useful inherent features were automatically extracted from historical TEC LSTM layers, performance proposed model was enhanced considering solar flux geomagnetic activity data. The ED-LSTME validated using 15-min values GPS measurements over one cycle (from January 2006 July 2018) collected at 15 stations in China. Different assessment experiments conducted different geographical locations seasons as well under varying activities, comprehensively evaluate model's performance. These comparative an ED-LSTM, traditional LSTM, deep autoregressive integrated moving average, 2016 International Reference Ionosphere models. results indicated that superior other statistical models, with R2 root mean square error 0.89 12.09 TECU, respectively. addition, adequately predicted conditions, satisfactory obtained even geomagnetically disturbed conditions. suggest prediction could be significantly improved utilizing auxiliary observations confirm outperforms several state-of-the-art models predictions times diverse