基于神经网络模型的紫外极光卵边界建模

Modeling of ultraviolet auroral oval boundaries based on neural network technology

极光卵的尺度大小和太阳风、磁层、电离层以及它们间的耦合过程有密切的联系,会随着空间和地磁环境的变化而变化.建立准确的极光卵边界预测模型对空间天气的预报以及了解日地关系具有重要意义.本文利用误差反向传播(back propagation, BP)神经网络和广义回归神经网络(general regression neural network, GRNN)两种神经网络模型对极光卵边界进行建模.结果显示GRNN的极光卵边界模型具有较高的准确性,赤道向边界预测平均绝对误差在0.77～1.20磁纬度(MLAT);极向边界预测平均绝对误差在0.83～1.39 MLAT.基于GRNN的极光卵边界模型预测准确性分别在极向边界和赤道向边界的整个磁地方时(MLT)上比BP神经网络的极光卵边界模型平均提高了0.74和0.73 MLAT,比多元线性回归模型平均提高了0.82和0.82 MLAT.而在模型的外推性方面, GRNN的极光卵边界模型的外推性优于BP神经网络的极光卵边界模型,与多元线性回归模型接近.

The size of auroral oval is closely related to the solar wind, magnetosphere, ionosphere and the coupling process between them. It is very important to establish an accurate boundary model for the space weather prediction and the understanding of the Solar-terrestrial interactions. In this paper, BP neural network and generalized regression neural network(GRNN) is used to construct auroral oval boundary model. The result shows that the auroral oval boundary predict model based on GRNN is more reliable, of which the mean absolute error in equatorial boundary is 0.77–1.20 magnetic latitude(MLAT), and the mean absolute error in poleward boundary is0.83–1.39 MLAT. Compared with the result of the BP network model, the accurancy of the GRNN model has improved 0.74 and 0.73 MLAT in poleward boundary and equad boundary respectly. Compared with the result of the multiple linear regression model, the accurancy of the GRNN model has improved 0.82 and 0.82 MLAT in poleward boundary and equatoward boundary respectly. In terms of the extrapolation of the model, the extrapolation of GRNN model is better than that of BP model, and is close to the multiple linear regression model.