EOF-LSTM神经网络的电离层TEC预报模型

为有效利用电离层总电子含量序列的时间信息，提出一种经验正交函数分解与长短期记忆神经网络组合的预报模型，利用IGS提供的云南地区TEC格网数据，分别对不同地点和不同时段的电离层进行建模预报。实验结果表明，该模型在同一时段预报5 d的TEC值均方根误差最优达1.83 TECu，较单一模型减小16%，其平均相对精度最优达91.56%，较单一模型增加7%；在同一地点预报5 d的TEC值均方根误差最优达1.86 TECu，较单一模型减小25%，其平均相对精度最优达90.74%，较单一模型增加7%。

Prediction Model of Ionospheric TEC by EOF and LSTM Neural Network

In order to effectively utilize the time information of the ionospheric total electron content (TEC) series, we propose a prediction model combining the empirical orthogonal function decomposition and the neural network of long and short memory. We use the TEC grid data of Yunnan region provided by IGS to model and forecast the ionosphere at different locations and different periods. The experimental results of the prediction model in this paper show that the optimal root mean square error of TEC values for 5 days in the same period of time is 1.83 TECu, which is reduced by 16% compared with the single model, and the optimal average relative accuracy is 91.56%, which is increased by 7% compared with the single model. The optimal root mean square error of TEC values at the same location for 5 days is 1.86 TECu, which is 25% less than that of the single model, and the optimal average relative accuracy is 90.74%, which is 7% more than that of the single model.