基于PCA-DNMFSC的卫星异常检测方法研究  

Fault detection for Satellite Telemetry based on PCA-DNMFSC

作　　者：彭艺[1] 冯小虎[1] 贾树泽[1] 韩琦[1] PENG Yi;FENG Xiao-hu;JIA Shu-ze;HAN Qi(National Satellite Meteorological Center,Beijing 100081,China)

机构地区：[1]国家卫星气象中心,北京100081

出　　处：《计算机仿真》2023年第1期48-52,142,共6页Computer Simulation

基　　金：国家重点研发项目(2018YFC1507803)。

摘　　要：为了实现卫星的智能化健康管理,提出基于主成分分析-动态稀疏化非负矩阵分解(PCA-DNMFSC)进行卫星遥测异常的自动检测。DNMFSC将卫星遥测数据分解成基向量,并使基矩阵稀疏化,致使产生异常的特征凸显,从而实现异常的检测。考虑到卫星遥测数据时序相关性,提出样本数据基于前l时刻的观测数据进行动态化表示;考虑到DNMFSC对基矩阵和系数矩阵的初始化是随机的,影响算法稳定性,采用主成分分析法(PCA)对DNMFSC进行初始化处理;通过构建的统计量的累计贡献率确定异常由哪些变量产生,从而识别异常。通过不同卫星的实际数据进行实验验证,结果表明利用正常的观测数据,可以实时检测卫星遥测数据出现的异常,有效避免故障漏报。In order to realize the intelligent health management of satellites, this paper proposes a Principal Components Analysis-Dynamic Non-negative Matrix Factorization with Sparseness Constraints(PCA-DNMFSC) method to automatically detect the abnormality of satellite telemetry. DNMFSC method decomposes the satellite telemetry data into the basis vector and makes it sparse. This benefits the prominence of characteristics of the abnormality, which makes the detection of fault easier. Considering the correlation of satellite telemetry data sequence, the paper proposes is a dynamic representation method of sample data. The initialization of DNMFSC is random in two aspects, one is the coefficient matrix, and the other is the base matrix, which will affect the algorithm stability. Principal Component Analysis(PCA) is put forward to initialize the DNMFSC algorithm. The effectiveness of the method is verified by the measured data of meteorological satellites. The cumulative contribution rate of the constructed statistics is used to determine which variables produce the anomaly, and then identify the anomaly. The results show that the observed data under normal conditions can be used to detect abnormal satellite telemetry, and the fault omission can be avoided effectively.

关 键 词：卫星 异常检测 累计贡献率 动态稀疏化非负矩阵分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]