| 引用本文: | 王婵,王慧泉,金仲和.皮纳卫星遥测数据异常检测聚类分析方法[J].哈尔滨工业大学学报,2018,50(4):110.DOI:10.11918/j.issn.0367-6234.201612148 |
| WANG Chan,WANG Huiquan,JIN Zhonghe.Pico-satellite telemetry anomaly detection through clustering[J].Journal of Harbin Institute of Technology,2018,50(4):110.DOI:10.11918/j.issn.0367-6234.201612148 |
|
| 摘要: |
| 为满足皮纳卫星高维遥测数据的实时、自动化、抗概念漂移等处理要求, 提出一种基于聚类的遥测数据异常检测方法, 包括子空间搜索和两阶段遥测数据聚类处理两部分.子空间搜索, 通过熵值实现所有遥测数据低维子空间划分, 降低计算复杂度, 避免“维度灾难”的发生; 两阶段遥测数据聚类处理, 在线阶段通过网格索引实时发现单点异常, 离线阶段通过聚类挖掘数据的集体异常及其特征, 满足快速异常检测和复杂异常检测两种需求, 并通过正常状态数据的迭代更新和算法的自适应修改, 抵抗概念漂移.ZDPS-1A卫星历史遥测数据的分析结果表明, 皮纳卫星遥测数据异常检测聚类方法在线阶段能实时处理10 kHz的流量数据, 发现95%的单点异常, 满足皮纳卫星实时遥测数据异常检测的一般需求; 算法自适应了卫星快速转动导致的数据漂移, 维持了稳定的单簇形态; 同时相比原边界检查系统早一个月检测出姿态确定与控制系统中程序跑飞引起的太阳敏感器数据紊乱故障.所提出的算法针对性解决了高维、存在概念漂移的遥测数据异常检测问题, 能实时检测单点异常, 具有集体异常挖掘能力, 适用于皮纳卫星星座组网的地面监控系统.
|
| 关键词: 皮纳卫星 异常检测 遥测数据 聚类 在线检测 子空间搜索 |
| DOI：10.11918/j.issn.0367-6234.201612148 |
| 分类号:V474 |
| 文献标识码:A |
| 基金项目:MEMS与皮卫星项目(61525403)；装备预研教育部联合基金 |
|
| Pico-satellite telemetry anomaly detection through clustering |
|
WANG Chan,WANG Huiquan,JIN Zhonghe
|
|
(Micro-satellite Research Center, Zhejiiang University, Hangzhou 310027, China)
|
| Abstract: |
| To meet the real time, automation, anti-concept drift and other processing requirements of pico-satellite high dimensionality telemetry data, a new cluster-based telemetry data anomaly detection method is proposed.The method contains correlative subspace search and two-phase density-based clustering processing for telemetry data.Subspace search calculates the entropy and entropy gain value to achieve low-dimensional subspace partitions of all telemetry data, to reduce the computational complexity and to avoid the occurrence of "dimension disasters".Two-phase telemetry data clustering algorithm meets the requirements of rapid anomaly detection and complex anomaly detection.In the online phase, the grid indexes are used to find single anomalies in real time, in the off-line phase, their specific features of the selected mining data are clustered to find collective anomalies.It also supports the anti-concept drift by iterative updating of normal data features and adaptive modification of the algorithm.The analysis for telemetry data of ZDPS-1A shows that in the on-line phase 10 kHz flowing data can be processed in real time and 95% of single-point anomalies can be found to meet the general needs of real-time telemetry data anomaly detection of Pico Satellite.This method adapts itself to change satellite conditions.The single cluster stays stable when the satellite rotates faster.This approach also detects an anomaly caused by attitude determination and control system's breakdown one month earlier than the current limit-checking system.The proposed algorithm solves the anomaly detection problem of telemetry data with high dimension and concept drift.It is suitable for the ground monitoring system of pico-satellite constellations.
|
| Key words: pico-satellite anomaly detection telemetry data clustering online monitoring subspace searching |
