基于Bi-LSTM神经网络的大气阻力系数预测

Prediction of Atmospheric Drag Coefficient Based on Bi-LSTM Neural Network

大气阻力难以精确建模,是低轨卫星精密定轨与轨道预测最大的动力学误差源。定轨处理时考虑利用时变的大气阻力系数(C_(D))来吸收大气阻力模型误差,从而获得较好的轨道拟合结果。然而由于缺少精确的建模方法来反映C_(D)参数的时变特征,导致轨道预报误差逐渐发散。针对该问题,提出基于Bi-LSTM神经网络预测C_(D)参数的轨道预报方法。首先通过动力学定轨方法解算GRACE-C卫星(GRCC)和Sentinel-3A卫星(SN3A)长期的C_(D)参数序列,然后采用Bi-LSTM神经网络方法进行C_(D)参数预测。结果显示,GRCC和SN3A卫星C_(D)预测值的MAE均值分别为0.0302和0.0774,RMSE均值分别为0.0416和0.1018。将C_(D)参数预测结果运用到两颗卫星4组轨道预报实验中,结果表明,GRCC卫星预报7d的最高平均精度为12.28m,平均精度提升率均在90%以上;SN3A卫星最高平均精度为16.00m,平均精度提升率最高可达74.82%。

It is difficult to accurately model atmospheric drag,which is the biggest error source of low-orbit satellite precise orbit determination and orbit prediction.In orbit determination,the time-varying atmospheric drag coefficient(C_(D))is considered to absorb errors of atmospheric drag model,obtaining better orbit fitting results.However,the time-varying characteristics of C_(D)parameter lack accurate modeling methods,resulting in gradually divergent orbit prediction errors.To solve this problem,we propose an orbit forecasting method for predicting C_(D)parameter based on Bi-LSTM neural network.Firstly,we calculate the C_(D)parameter of GRACE-C and Sentinel-3A satellites by dynamic orbit determination method.Then,we predict C_(D)parameter by Bi-LSTM neural network.The results indicate that the mean MAE values of C_(D)parameters predicted by GRCC and SN3A satellites are 0.0302 and 0.0774,and the mean RMSE values are 0.0416 and 0.1018.The results of four groups of orbit prediction experiments show that the highest average accuracy of GRCC satellite is 12.28 meters,and the average accuracy improvement rate is above 90%.The highest average accuracy of SN3A satellite is 16.00 meters,and the average accuracy improvement rate is up to 74.82%.