一种电离层TEC格点预测模型

基于分析时间序列数据的门限控制单元（GRU）神经网络模型，利用电离层TEC网格点历史数据、太阳活动指数、地磁活动指数作为预测因子，提出一种高精度电离层TEC格点预测模型.对全球60个网格点的数据进行了模型预测和对比实验，得到北半球平均相对精度的均值为83.96%，高于南半球的73.60%，表明预测模型在北半球的适应性更好，且中低纬地区的适应性优于高纬地区；预测模型在磁扰动期的平均相对精度的均值比磁平静期平均相对精度的均值高，约1.95%；与基于递归神经网络（RNN）、长短时记忆网络（LSTM）和双向长短时记忆网络（Bi-LSTM）的电离层TEC单站预测模型相比，本文预测模型的均方根误差（RMSE）平均为原来的80.8%. 

A Prediction Model of the Grid Point Ionospheric TEC

A high-precision ionospheric TEC grid point prediction model is established using Gate Recurrent Unit (GRU) neural network model suitable for analyzing time series data. Ionospheric TEC grid point historical data, solar activity index, and geomagnetic activity index are used as inputs of our model. After our in-depth research and analysis, the data of 60 grid points were employed to predict model and do comparative experiments, and the results show that the mean value of average relative accuracy of the northern hemisphere is 83.96%, higher than 73.60% of the southern hemisphere. It presents that the adaptability of the prediction model is better in the northern hemisphere, and especially in the middle and low latitudes rather than in the high latitudes. The second result is that the mean value of the average relative accuracy of the prediction model in magnetic disturbance period is about higher 1.95% higher than that in magnetic quiet period. Finally, we compared the prediction results of several representative models. Compared with the single station prediction model based on RNN, LSTM and Bi-LSTM, the RMSE of this prediction model is reduced to 80.8% on average.