SVM回归估计方法在卫星故障诊断中的应用  

Application of SVM regression to fault diagnosis for satellite

作　　者：赵石磊[1] 张迎春[1] 

机构地区：[1]哈尔滨工业大学航天学院,黑龙江哈尔滨150001

出　　处：《电机与控制学报》2008年第4期483-486,共4页Electric Machines and Control

摘　　要：针对红外地平仪、陀螺组成的卫星姿态测量系统,研究了系统中任一姿态敏感器的故障诊断问题。首先采用Lanckriet的半正定规划方法对支持向量机核搜索空间进行优化,利用此支持向量机的回归估计方法对卫星姿态测量系统的输入输出历史数据进行学习,进而拟合得到系统输入输出间的冗余关系,并将这种对应关系应用于卫星姿态测量系统传感器的故障检测与诊断中。仿真结果表明,该方法能够正确地估计系统输入输出关系,有效地提取故障信息,而且计算过程相对简单,易于实现。This paper studied a kind of fault diagnosis problems in the satellite attitude measurement system composed of gyroscopes and infrared earth sensors, when there was only one of sensors in fault. Firstly, it used Lanckriet＇ s Semi-Definite programming method to optimize the searching space of the SVM （Support Vector Machine） kernel. Then, this SVM regression method was used to learn the historical datasets of output and input to acquire the redundancy relation of the system, and this relation was applied to the fault detection and diagnosis of the sensors of the SAMS. Simulation result shows that this method could estimate these relations of the system correctly and acquire fault information effectively, while this method can lighten the computation burden and is easy to be actualized.

关 键 词：支持向量机 姿态测量系统 故障诊断 

分 类 号：V448.25[航空宇航科学与技术—飞行器设计] TP277[自动化与计算机技术—检测技术与自动化装置]