Research On Satellite Anomaly Detection Technology Based On Multivariate Correlation Analysis

During the operation of the satellite,the ground receives the remote telemetry data and performs satellite operation management and status monitoring based on this data.The detection of potential anomalies in the telemetry data in the satellite's early failure is of great significance to the safety maintenance of the satellite.The object of satellite anomaly detection is a large number of telemetry parameters.The telemetry parameters have the characteristics of multidimensionality,correlation,and nonlinearity.According to the characteristics of satellite telemetry data and the type of satellite failures,the existing algorithms are improved and applied to satellite anomalies detecting field.First,it describes the characteristics of satellite telemetry parameters and satellite failures,and summarizes the main method system of satellite anomaly detection,and the advantages and disadvantages of various methods.On this basis,the existing multivariate data analysis and processing methods,including variable screening and dimensionality reduction,are analyzed,and the advantages and disadvantages of various methods are summarized,which lays a theoretical foundation for satellite anomaly detection methods.Secondly,a PCA anomaly detection method based on the correlation probability model is proposed.The data is divided according to the satellite's duty cycle,and the change of the correlation between the telemetry parameters is studied in every work unit,which is more convincing than detecting a single outlier.PCA is used to reduce the dimensionality of the multivariate probability model.In addition,the~2 statistic is used to judge whether the data is abnormal,which avoids the trouble of threshold setting.In addition,after detecting abnormal data,the abnormal telemetry parameters can be determined by calculating the reconstruction error contribution ratio.It is verified by simulation that the method can detect anomalies in the early failure,and the experimental results are compared and analyzed.Moreover,this method can quickly help the transportation management people to diagnose the early failure,so that the anomaly can be processed in time to avoid the bigger accident.Finally,for the use of traditional GPR models for high-dimensional satellite telemetry data,which results in reduced prediction performance,over-fitting of the training model and high false alarm,a satellite anomaly detection method based on the DC-GPR model is designed.It selects prediction variables by distance correlation coefficients to improve the interpretability of the model,estimates the generalization error of the model to set a more reasonable prediction interval and improve the generalization ability of the model,reduce the false alarm,and the prediction error can intuitively reflect the abnormal degree of the data.Simulation experiments were carried out on the actual satellite data and compared with other methods.It was verified that this method can detect data anomalies in the satellite's early failure,make the prediction model more superior and reduce the false alarm.