Research And Application Of Anomaly Detection Method For Telemetry Data Of Space Information Networks

The Space Information Networks(SINs)is a complex network system capable of acquiring,processing,storing and transmitting space information in real time,which expands the space of human activities to deep space and even deep sea,and play an indispensable role in a number of fields such as national development and livelihood,resource exploration,scientific investigation and national defense security,and have become a highlight for industrial development and global science and technology.SINs contain a large number of spacecraft that complement the ground networks in terms of coverage and mobile access,etc.These spacecraft help the space information network realize integration applications and collaborative services based on satellite remote sensing,satellite navigation and satellite communications.However,due to the diversity of external factors and performance degradation,spacecraft failures or anomalies occur occasionally.The anomalies generated by spacecraft systems and components are characterized by complex causes,unpredictable duration,enormous losses and serious potential problems.Therefore,it is vital to study the anomaly detection methods of spacecraft telemetry data in SINs and develop anomaly detection systems to enhance spacecraft health management and guarantee SINs’ normal operation.Aiming at the characteristics of telemetry data of SINs,this thesis focuses on the existing problems of telemetry data anomaly detection,thoroughly studies the anomaly detection technology of telemetry data of SINs,proposes a telemetry data anomaly detection method based on the temporal dependence and intrinsic associations between sequence,and designs and implements a SINs telemetry data anomaly detection system based on the proposed method.To summarize,the innovations and major contributions of this thesis are listed below.(1)To address the problem that there are complex dependencies between sequences,yet the collected data lack priori information about the graph structure,this thesis proposes a graph-based multidimensional data correlation discovery method.This method discovers deterministic correlations that can be expressed by formulas as well as uncertain correlations that cannot be expressed by formulas or functions and further learns the multivariate time series graph structure to help subsequent anomaly detection.Experiments demonstrate that the method is able to learn effective dependencies between multivariate time series.(2)Telemetry data are multivariate time series data with both spatial and temporal correlation in continuous space,but the existing telemetry data anomaly detection methods fail to make full use of various complex dependencies in both temporal and spatial dimensions of telemetry data for anomaly detection.To address this problem,on the basis of work(1),this thesis proposes a method for anomaly detection of telemetry data in SINs using temporal and spatial information.The method fully considers the characteristics of telemetry data in order to give more accurate detection results for anomaly detection tasks in the aerospace field.Experiments prove that the method has the ability to accurately detect and precisely locate anomalies and provide explanations for the detected anomalies.(3)Based on the proposed SINs telemetry data anomaly detection method in this thesis,a SINs telemetry data anomaly detection system is designed and implemented.