Fault detection for satellite attitude sensors are required to handle high-dimensionality telemetry data and to manipulate faults in real-time. A new method based on exact Euclidean locality-sensitive hashing and subspace outlier degree is proposed to satisfy aforementioned requirements. The algorithm detects the global abnormal points by establis-hing and using the exact Euclidean local sensitive hash index. The subspace abnormal points are detected after calculat-ing the subspace outlier factor. This paper presents the new concepts of approximate reference points and cache bucket to reduce the time complexity. The analysis for ZDPS-2 satellite attitude sensor data shows that,the precision is 89.3%and the recall is 100%. In addition,the algorithm's generalization ability is superior to the original subspace outlier degree al-gorithm. The proposed method solves two problems of the original subspace outlier degree. It can satisfy the real-time and global abnormal points manipulation requirements of attitude sensors for the attitude control system.%为满足高维、多状态姿控敏感器遥测数据的实时故障检测,提出了一种基于局部敏感哈希和子空间异常因子的故障检测算法.算法通过局部敏感哈希索引的建立和使用,检测全局故障点;通过子空间异常因子的计算,检测子空间故障点.提出了近似邻近参考集与缓存桶的概念,降低算法的时间复杂度.ZDPS-2卫星的姿控敏感器数据分析结果表明,该方法故障查准率89.3%,查全率100%,且泛化性能优于原始的子空间异常程度算法.该算法解决了原始的子空间异常程度算法实时性低、检测全局故障困难问题,可以满足姿控敏感器实时故障检测需求.
展开▼
