摘要
飞行器遥测数据是地面判断卫星在轨状态的唯一来源。异常检测有助于飞行器运行过程的视情动态决策,并能有效减少故障。然而,现有方法主要关注短时变化,难以有效识别集合异常模式。针对这一问题,提出了一种基于长时间尺度特性建模优化的飞行器遥测数据集合异常检测方法。首先,构建时序关联依赖模型,提取遥测数据片段中的高维时序规律并生成预测结果;然后,利用预测结果与观测数据之间的残差,构建统计模型,提取分布特征并形成异常检测判据;最后,利用迭代预测自动调整模型输入,提升集合异常检测的鲁棒性。通过实际飞行器姿态角数据的验证,结果表明,相比VAE-LSTM模型,异常片段的检出率提升了0.041,F1分数提升了0.039,证明了该方法在提高检测精度和降低漏检率方面的优势,为卫星视情运维提供可靠的基础数据支撑。
Satellite telemetry data is the sole source for ground-based assessment of satellite in-orbit status.Anomaly detection facilitates condition-based dynamic decision-making during satellite operations and effectively reduces failures.However,existing methods primarily focus on short-term variations,making it challenging to identify collective anomaly patterns effectively.To address this issue,this paper proposes a collective anomaly detection method for satellite telemetry data based on long time-scale characteristic modeling optimization.First,a temporal correlation model is constructed to extract high-dimensional patterns from telemetry data segments and generate prediction results.Then,using the residuals between the prediction results and observed data,a statistical model is developed to extract distribution characteristics and establish anomaly detection criteria.Finally,iterative prediction is employed to automatically adjust model inputs,enhancing the robustness of collective anomaly detection.Validation using actual satellite attitude angle telemetry data shows that,compared with the VAE-LSTM model,the proposed method improves the detection rate of anomaly segments by 0.041 and the F1 score by 0.039.These results demonstrate the method′s advantages in improving detection accuracy and reducing missed detections,providing reliable data support for condition-based satellite operations and maintenance.Aircraft telemetry data are the only source for ground-based assessment of satellite in-orbit status.Anomaly detection facilitates condition-based dynamic decision-making during aircraft operations and effectively reduces failures.However,existing methods primarily focus on short-term variations,making it difficult to identify collective anomaly patterns effectively.To address this issue,this article proposes a collective anomaly detection method for aircraft telemetry data based on long time-scale characteristic modeling optimization.First,a temporal correlation model is formulated to extract high-dimensional patterns from telemetry data segments and generate prediction results.Then,using the residuals between the prediction results and observed data,a statistical model is developed to extract distribution characteristics and establish anomaly detection criteria.Finally,iterative prediction is employed to automatically adjust model inputs,enhancing the robustness of collective anomaly detection.Validation using actual aircraft attitude angle telemetry data shows that,compared with the VAE-LSTM model,the proposed method improves the detection rate of anomaly segments by 0.041 and the F1 score by 0.039.These results show the method′s advantages in improving detection accuracy and reducing missed detections,providing reliable data support for condition-based satellite operations and maintenance.
作者
孙家正
宋宇晨
崔展博
李桢煜
王智鹏
刘大同
Sun Jiazheng;Song Yuchen;Cui Zhanbo;Li Zhenyu;Wang Zhipeng;Liu Datong(School of Electronics and Information Engineering,Harbin Institute of Technology,Harbin 150080,China;Shijiazhuang Hai Shan Aviation Electronic Technology Company Ltd.,Shijiazhuang 050200,China;AVIC Xi’an Flight Automatic Control Research Institute,Xi’an 710076,China)
出处
《仪器仪表学报》
CSCD
北大核心
2024年第11期312-321,共10页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金青年项目(62201177)项目资助。
关键词
飞行器
遥测数据
异常检测
长时预测
aircraft
telemetry data
anomaly detection
long-term prediction
作者简介
孙家正,2019年于哈尔滨工业大学获得学士学位,现为哈尔滨工业大学自动化测试与控制研究所硕士研究生,主要研究方向为卫星数据智能挖掘、飞行器健康状态智能评估等。E-mail:hit_sunjiazheng@163.com;通信作者:宋宇晨,2015年于电子科技大学获得学士学位,2022年于哈尔滨工业大学获得博士学位,现为哈尔滨工业大学副教授,主要研究方向为复杂系统智能感知与评估、智能测试信息处理、卫星星座状态监测与运维等。E-mail:songyuchen@hit.edu.cn。
