2015年3月特大磁暴期间中国区域电离层TEC NeuralProphet预报模型研究

延迟是全球卫星导航定位中重要的误差源之一，提高电离层TEC建模和预报精度对改善卫星导航定位精度至关重要.本文构建了以太阳辐射通量指数F_10.7、地磁活动指数Dst、地理坐标和中国科学院(Chinese Academy of Sciences，CAS) GIM数据为输入参数的NeuralProphet神经网络模型(NP模型)，实现在2015年3月特大磁暴期中国区域电离层TEC短期预报.为验证NP模型的预报精度，本文同时构建了长短期记忆神经网络(Long Short-term Memory Neural Network，LSTM)模型进行对比分析.结果统计分析表明，NP模型在磁暴期(2015年DOY076-078) TEC预报值RMSE和RD分别为0.83 TECU和3.13%，绝对和相对精度较LSTM模型分别提高1.49 TECU和10.25%；且NP模型RMSE优于1.5 TECU的比例达97.24%，远高于LSTM模型.NP模型预报值与CAS具有较好一致性和无偏性，偏差均值仅为-0.01 TECU，而LSTM模型预报值的均值偏大，偏差均值为1.49 TECU.从低纬到中纬度的三个纬度带内，NP模型RMSE分别为1.12、0.83和0.44 TECU，精度比LSTM模型提高1.94、1.56和1.23 TECU.整体上，在磁暴期NP模型预报性能明显优于LSTM模型，能够精细描述中国区域电离层TEC时空变化.

Study on the NeuralProphet forecast TEC model over China during the severe geomagnetic storm in March 2015

Ionospheric delay is one of the significant error sources in global navigation satellite system. It is essential to improve the accuracy of ionospheric TEC modeling and forecasting for enhancing the accuracy of satellite navigation positioning. In this paper, a NeuralProphet neural network model (NP) is constructed with solar radiation flux index (F10.7), geomagnetic activity index (Dst), geographic coordinates and GIM from Chinese Academy of Sciences (CAS) as influence factors and input parameters. And the presented NP model is applied for the short-term forcasting of ionospheric TEC over China during the severe magnetic storm in March 2015. In order to verify the performence of NP model, a Long Short-term Memory Neural Network (LSTM) model is implemented for comparative analysis. The statistical analysis results show that the root mean square error (RMSE) and relative deviation (RD) of NP model during the geomagnetic storm period (DOY076-078) are 0.83 TECU and 3.13%, respectively, which are 1.49 TECU and 10.25% more accurate than LSTM model from the perspectives of absolute and relative accuracy. And for the ratio of RMSE less than 1.5 TECU, NP forecast model is about 97.24%, which is much better than LSTM model. The TEC predictions from NP model has good consistency and unbiasedness with CAS-TEC showing the mean bias of -0.01 TECU, while the LSTM model has a larger mean bias of 1.49 TECU. The RMSE of NP model are 1.12, 0.83 and 0.44 TECU, respectively, from low to mid-latitudinal zone, and the forecasting accuracy is 1.94, 1.56 and 1.23 TECU higher than LSTM model. The proposed NP forecast model has a significantly better forecasting performance than LSTM, which would be useful for characterizing the spatial-temporal characteristics more accurately under disturbed conditions over China.