基于BP神经网络的GEO等离子体环境参数反演分析

空间等离子体环境诱发的表面充电效应会对航天器运行产生干扰，严重时将导致太阳电池等部件失效。通过神经网络反演方法，以GEO环境中介质表面充电电位曲线作为输入，在双峰麦克斯韦分布假设下，可以逆向得到高能峰的等离子体参数。分析了GEO等离子体环境参数对表面充电电位曲线的影响，表明高能峰在充电过程中起决定性作用；其次通过MATLAB搭建BP神经网络，采用COMSOL计算得到多组充电曲线进行网络训练和反演计算，得到等离子体密度反演的平均相对误差为0.42%，温度反演的平均相对误差为0.03%，整体误差在0.1%～5.6%。结果表明，采用神经网络对等离子体环境进行反演具有可行性，该方法可以作为空间等离子体环境探测结果的对比参考和航天器非探测点表面电位计算的输入条件。

Inversion Analysis of GEO Plasma Environmental Parameters Based on BP Neural Network

The surface charging effect induced by the space plasma environment may interfere with the operation of spacecraft, which can lead to the failure of solar cells and other components. It is shown that the high-energy plasma environmental parameters can also be obtained by taking the electrical potential curve of the dielectric surface in the GEO environment as input under the assumption of double Maxwell distribution. Firstly, the influence of GEO plasma environment parameters on the surface charging potential curve is analyzed, indicating that the high energy peak plays a decisive role in the charging process. Then, BP neural network is established by MATLAB, and the network training data is obtained according to multiple groups of charging curves calculated by COMSOL. Finally, the average relative error of plasma density and temperature inversion is 0.42% and 0.03%, and respectively, and the overall relative error is within 0.1%~5.6%. Results showed that it was feasible to use a neural network to invert the plasma environment, and the method can be used as a reference for comparison of the detection results of the space plasma environment. The inversion results can be used as the input conditions for surface potential calculation of non-detection points of spacecraft.