摘要
针对北斗某星辐射剂量探测数据缺失问题,提出了一种基于线性样条和CNN-LSTM神经网络模型的处理方法。在对数据特性分析的基础上,将原始数据分解为线性趋势项和季节波动项。对于线性趋势项,采用基于线性样条的缺失值处理方法;对于季节波动项,根据其时空变化特性,设计CNN和LSTM组合神经网络结构,完成季节波动项的缺失值处理。实验表明,相比于线性插值法和傅里叶变换插值方法,本文所提方法的插补值与真实值偏差更小,相关性更高。平均相对误差达到0.008,相关系数达到0.855。同时横向对比了本文所提组合神经网络模型和单一的LSTM和CNN网络模型的插补结果,同样本文方法表现出更好的一致性。研究结果表明,本文方法能够较好解决北斗数据连续缺失的问题,为后续基于北斗数据开展科学研究和业务应用奠定基础。
Aiming at the problem of missing radiation dose detection data of a certain Beidou satellites, a processing method based on linear spline and CNN-LSTM is proposed. First, analyze the characteristics of the data and decompose it into long-term trend items, a linear spline method is used to complete the missing value processing. For the seasonal fluctuation items, the temporal and spatial change rules are analyzed, and on this basis, the CNN and LSTM fusion neural network structure is designed to complete the processing of the seasonal fluctuation items. Experiments show that, comapred with the linear interplation method and the Fourier transform method, our approach performs better in the prediction with an average relative error of 0.008 and a correlation coefficient of 0.855. At the same time, the prediction accuracies of our approach and the single LSTM and single CNN network models are compared, similarly the predictions of our approach is more consistent with the observed data, and has a smaller deviation. The results show that the approach in this paper can better solve the problem of continuous missing values of Beidou satellites observations, which lays a foundation for the further scientific research based on the dataset.
作者
杨旭
崔瑞飞
田超
胡斯惠
姜健民
徐培康
YANG Xu;CUI Ruifei;TIAN Chao;HU Sihui;JIANG Jianmin;XU Peikang(32035 Unit of PLA,Xi’an 710600;College of Meteorology and Oceanography,National University of Defense Technology,Changsha 410073)
出处
《空间科学学报》
CAS
CSCD
北大核心
2022年第1期163-169,共7页
Chinese Journal of Space Science
基金
国防科技创新特区项目资助(1916321TS00101206)。
