一种数据驱动的卫星健康状态判读模型  

A Data-driven Model for Satellite Health State Interpretation

作　　者：刘传鲁 李常亮 高伊萱 章雷 薛彬 LIU Chuanlu;LI Changliang;GAO Yixuan;ZHANG Lei;XUE Bin(Aerospace Science and Industry Space Engineering Development Co.,Ltd.,Beijing 100854,China)

机构地区：[1]航天科工空间工程发展有限公司,北京100854

出　　处：《航天器工程》2024年第1期123-130,共8页Spacecraft Engineering

摘　　要：针对海量卫星健康管理数据的处理面临参数维度高、冗余参数多、参数间关系难量化、健康状态难以定量判读的问题,文章提出一种结合相关性聚类分析的数据处理机制和多分类器集成的健康状态判读模型。首先通过最大信息系数(MIC)量化参数关系并提取关键特征,然后应用聚类分析将关键特征参数数据转换成健康状态知识库,最后基于健康知识库集成训练多分类器来实时监测卫星的健康状态。应用某卫星载荷的遥测数据对该模型有效性进行了验证,结果表明:该判读模型经过数据挖掘后的关键关联参数训练,具有较好的卫星异常状态识别能力,其准确度达到了98%,可为在轨卫星健康状态监视手段的选择提供参考。In response to the problems faced by the processing of massive satellite health management data,such as high parameter dimensions,multiple redundant parameters,difficulty in quantifying the relationship between parameters,and difficulty in quantitatively interpreting the health status,a data processing mechanism combing correlation cluster analysis and a health state interpretation model integrating multi-classifier integration is proposed in this paper.Firstly,the maximum information coefficient(MIC)is used to quantify the parameter relationships and the key feature parameters are selected.Then the key feature parameter data is converted into the health status knowledge database by cluster analysis.Finally,multiple classifiers are trained based on the health knowledge database to monitor the satellite health status in real-time.The validity of the model is verified using the telemetry data from a certain satellite payload.The simulation results show that the model trained with key associated parameters after data mining,has good ability to detect satellite abnormal state with an accuracy of 98% it can provide reference for the selection of on-orbit satellite health monitoring means.

关 键 词：最大信息系数 特征选择 聚类 多分类器集成 异常检测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]