基于深度学习的SuperDARN雷达极区电离层跨极盖电场模型构建

MODELING OF THE SUPERDARN POLAR IONOSPHERIC CROSS POLAR CAP ELECTRIC FIELD USING DEEP LEARNING

通过超级双子极光雷达网(SuperDARN)获得的跨极盖电势计算了极区电离层对流电场。利用2014年的极区电离层对流电场数据为基础,引入对流电场的历史数据,分别基于多元线性回归算法和后向传播神经网络算法构建电离层电场模型。利用独立的数据集,验证了两种模型的准确性和稳定性。结果表明,模型值与测量值的均方根误差在2.0~3.5 mV·m^(-1)之间,平均绝对误差范围为1.5~3.0 mV·m^(-1),线性相关系数均大于0.6,最高可达0.9。引入前20分钟的历史数据作为模型的输入,后向传播神经网络模型比多元线性回归模型具有更好的预测性能。

The Cross Polar Cap electric field (i.e. the ionospheric convective electric field) was calculated using ionospheric potential data (i.e. the Cross Polar Cap potential) from the Super Dual Auroral Radar Network (SuperDARN). Then, historical data for the convective electric field were introduced and the ionospheric electric field model was constructed using a multivariate linear regression algorithm and a back propagation neural network algorithm that used ionospheric electric field data from 2014. The accuracy and stability of the two models were verified using an independent dataset. The results show that the root mean square of the error between the model values and the measured values is in the range of 2.0 mV·m^(-1) to 3.5 mV·m^(-1), the mean absolute error is in the range of 1.5 mV·m^(-1) to 3.0 mV·m^(-1), and the linear correlation coefficient is greater than 0.6 and has a maximum of 0.9. At the same time, the historical data of convection electric field in the first 20 min were introduced as the input for the multivariate linear regression model and BP neural network model. These results show that the back propagation neural network model has better prediction performance than the multivariate linear regression model.