论文部分内容阅读
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。
In order to meet the real-time fault detection of high-dimensional and multi-state attitude sensor telemetry data, a fault detection algorithm based on local sensitive hash and subspace anomaly is proposed. The algorithm detects global failure points through the establishment and use of local sensitive hash indexes. The subspace fault points are detected by calculating the abnormality factors of subspaces. The concept of approximate neighbor reference sets and cache buckets is proposed to reduce the time complexity of the algorithm. The result of ZADS-2 satellite attitude-sensor data analysis shows that the method has the accuracy of 89.3% and the check-up rate of 100%, and its generalization performance is superior to the original algorithm of subspace anomaly. The algorithm solves the problem of low real-time of the original subspace anomaly algorithm and the difficulty of detecting the global fault, and meets the requirements of the real-time fault detection of the attitude control sensor.