导航卫星载荷分系统健康评估方法设计

Design of Navigation Satellite Payload Subsystem Health Evaluation Method

摘要:

面向下一代导航系统结合高中低轨构建导航星座的设想,随着遥测参数数量和种类的激增,针对传统健康评估方法面临的过往专家知识难以适用、故障机理储备难以覆盖全面的问题,提出了基于局部异常因子检测-贝叶斯网络结构学习的导航卫星载荷分系统健康评估方法；通过采集某卫星系统实际故障时间点前后数据,设计实验验证了局部异常因子检测方法能够以粗粒度正确输出单机级的健康状况；分析比较了三种评分函数下,贝叶斯结构学习的效率和模型的准确度；实验结果表明,当评分函数分别选为BDeuScore、 K2Score以及BicScore时,学习到的模型对系统的健康评估准确度分别为87.4%、80.5%和85.2%；总结了局部异常因子检测-贝叶斯网络结构学习方法各自的不足,为导航卫星分系统健康评估方法提供了新方向和思路。

Abstract:

Envisioning the construction of navigation constellations combining high, medium, and low orbits for the next generation navigation systems, the exponential increase in the quantity and types of telemetry parameters poses challenges to traditional health evaluation methods. These challenges stem from the difficulty in applying past expert knowledge and the inadequacy of fault mechanism reservoirs for comprehensive coverage. To address this, a method for evaluating the health of navigation satellite payload subsystems based on Local Outlier Factor Detection and Bayesian Network Structure Learning is proposed. Experimentation involving the collection of data before and after actual fault occurrences in a satellite system validated the ability of the Local Outlier Factor Detection method to accurately output the health status at a coarse-grained level. Efficiency and accuracy of the Bayesian Network Structure Learning were analyzed and compared using three scoring functions. Experimental results indicated that when employing BDeuScore, K2Score, and BicScore as scoring functions, the learned models achieved respective accuracy levels of 87.4%, 80.5%, and 85.2% in assessing the system"s health. Limitations of the Local Outlier Factor Detection and Bayesian Network Structure Learning methods were summarized, providing new directions and insights for the health assessment of navigation satellite subsystems.