基于隐空间插值自编码器的卫星遥测参数异常检测  

Anomaly Detection of Satellite Telemetry Data Based on Latent Space Interpolation Autoencoder

作　　者：周台春 郭国航 肖志刚 李虎 ZHOU Taichun;GUO Guohang;XIAO Zhigang;LI Hu(National Space Science Center,Chinese Academy of Sciences,Beijing 100190;University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]中国科学院国家空间科学中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《空间科学学报》2024年第6期1155-1165,共11页Chinese Journal of Space Science

基　　金：中国科学院战略性先导科技专项科学卫星任务运控技术项目资助(XDA15040100)。

摘　　要：卫星遥测参数是地面运管系统评估卫星在轨运行正常状态的关键指标,遥测参数异常检测对于保障卫星安全可靠运行和任务顺利执行至关重要.针对现有卫星遥测异常检测算法对参数特征提取存在区分度缺乏、有效异常决策信息提取不充分等问题,本文提出一种基于隐空间插值优化的异常检测方法,将隐空间优化约束后的自编码器的表示学习能力与核密度估计方法的密度估计能力相结合,有效地进行异常检测.采用量子科学卫星的真实遥测参数数据和公开数据集进行验证,其结果表明所提方法在真实遥测参数上比最优对比方法的Auc值和F1值分别提升了5.6%和5.8%.与其他异常检测算法相比,该方法有较强的正常和异常样本辨别能力,有效解决了特征缺乏区分性以及决策信息提取不充分的问题,同时具有良好的噪声抗干扰性和有效性.Satellite telemetry parameters are the critical indicators for the ground operation and man-agement system to assess the normal state of satellite operation in orbit,and anomaly detection of telemetry parameters is essential to guarantee the safe and reliable operation of satellites and the smooth execution of tasks.In response to the existing satellite telemetry anomaly detection algorithms for pa-rameter feature extraction there is a lack of differentiation,effective anomaly decision-making informa-tion is not sufficiently extracted and other problems,this paper proposes an anomaly detection method based on the optimization of latent space interpolation,the latent space optimization constraints after the self-coder’s representation learning ability and the density estimation ability of the Kernel Density Estimation(KDE)method are combined to effectively carry out the anomaly detection.Real telemetry parameter data from quantum science satellites and public datasets are used for validation,and the re-sults show that the proposed method improves the Auc and F1 values over the optimal comparison method by 5.6%and 5.8%,respectively,on real telemetry parameters.Compared with other anomaly de-tection algorithms,the proposed method has strong ability to discriminate normal and abnormal sam-ples,effectively solves the problems of lack of differentiation of features and insufficient extraction of de-cision information,and has good noise immunity and effectiveness.

关 键 词：科学卫星 遥测参数 自编码器 异常检测 隐空间优化 

分 类 号：V557.3[航空宇航科学与技术—人机与环境工程]