利用人工神经网络快速计算木星系磁坐标

在木星辐射带研究中，从地理坐标向磁坐标的准确转换是建模基础.以往的建模中，磁壳参数L值的计算基于磁偶极场假设，该方法精确度较差.结合最新的高精度木星磁场模型JRM09，本文提出基于磁力线追踪法的木星磁坐标计算方法，并分析其合理性和必要性.要求精确度较高时，磁力线追踪法计算耗时很长.本文在磁力线追踪法的基础上进行改进，提出基于人工神经网络的磁坐标快速计算方法.该方法包括分类器和拟合器.分类器基于Adaboost算法的BP神经网络，用于预测某地理坐标是否在内磁层，如果在内磁层，则用拟合器计算L值.拟合器采用遗传算法优化BP神经网络.结果表明，分类器的分类错误率在3%以内，而拟合器的预测误差在7%以内.以Juno号一圈探测轨道为例，利用神经网络的磁坐标计算法比磁力线追踪计算法速度快3个数量级以上.基于人工神经网络的磁坐标快速计算方法可用于未来木星辐射带的研究.

Fast Calculation of Magnetic Coordinates Using Artificial Neural Network in Jovian System

In the study of Jovian radiation belt, the accurate calculation of magnetic coordinates from geographic coordinates is the basis. In the previous study, the L-shell parameters are calculated based on the assumption of a dipole field. The accuracy of this method is low. In this study, an L-shell calculation method based on magnetic field line tracing method is presented. The necessity and rationality of this method are also shown. This method is very time-consuming at the high-resolution requirement. In addition, this method is advanced by combining with an Artificial Neural Network (ANN), which can calculation L-shell at a much faster speed. The ANN consists of a classifier and a predictor. The classifier is a BP neural network based on Adaboost algorithm and can identify if the coordinates are within inner magnetosphere. If so, the predictor is used to calculate the L-shell parameter. The predictor is a BP neural network optimized by genetic algorithm. The classification error rate of the classifier is within 3% and the error rate of the predictor is within 7%. In the calculation of the orbit of Juno spacecraft, the calculation speed of this neural network based method is about 3 orders higher than that based on the field line tracing method.