论文部分内容阅读
目前的卫星轨道预报主要依靠动力学模型。由于模型与实际卫星所在空间环境变化存在差异,故通过动力学模型预报的轨道与实际轨道的偏差较大;尤其低轨卫星,由于空间环境复杂多变,预报误差更大;利用深度学习的神经网络作为轨道预报的工具,通过对卫星轨道数据的训练学习,掌握数据之间隐含的关系预测未来数据;将深度学习的长短时记忆神经网络模型优化,并将预报数据与实际数据进行对比分析,将预报20 d的误差由之前最大值的300 km降低到5 km以下,提高了神经网络预报卫星轨道的精度。
The current satellite orbital forecast mainly depends on the dynamic model. Due to the difference of the space environment between the model and the actual satellite, the orbit predicted by the dynamic model has large deviation from the actual orbit; especially the low-orbit satellite, the forecast error is greater due to the complicated and changing space environment; As a tool for orbit prediction, the network can predict the future data through the training and learning of satellite orbit data, and grasp the implicit relationship between the data. The neural network model of long and short-term memory for depth learning is optimized, and the forecast data and actual data are compared , The forecast error of 20 days from the previous maximum of 300 km down to 5 km or less, improve the accuracy of neural network to forecast satellite orbit.