Research On Correlation Knowledge Discovery Method Of Spacecraft Telemetry Data

Telemetry data not only has huge amount,but also has the charactersistics of high dimension and strong non-linearity,which brings challenges into its correlation analysis.Through the research work on the characteristics analysis,category abstraction,measurement selection and method improvement of telemetry data correlation,some methods of correlation knowledge discovery suitable for telemetry data are proposed,which is of great significance to the integrated test and operation management of satellites.The main contents of this thesis include:1)A novel method of inductive telemetry data autocorrelation knowledge discovery is proposed after modeling 6 kinds of autocorrelation relations which are ubiquitous in telemetry field.The experimental results show that the proposed method can automatically construct the relevant knowledge which is in line with the design of satellite system,and based on these knowledge,some anomalies of telemetry data can be detected effectively.2)Based on the strong non-linearity and dynamics of the cross-correlation,and the large-scale and high-dimensional characteristics of telemetry data,this paper puts forward the evaluation criteria of correlation measure,which measures universality by coverage and novelty,and timeliness by average running time.Quantitative analysis and experiments verify the optimum measure is Maximum Correlation Coefficient?MIC?for cross-correlation analysis of telemetry data under this standard.3)A novel method based on improved maximal information coefficient?MIC?is proposed to discover correlations in massive telemetry data efficiently.First,Mini Batch K-Means clustering algorithm is used to discretize data in the precursor process,then mutual information between two variables under this partition is calculated and normalized by information entropy instead of maximal entropy.Finally,the maximum information coefficient is selected as the measure of variable correlation.Apply the method to the correlation analysis of the quantum satellite telemetry data,results show that proposed method can effectively solve the problem of MIC measure bias to multi-valued variables compared with the method based on dynamic programming algorithm,the time complexity dropped from?9?^2.4)to?9?^1.6),and it is an effective method for large-scale data correlation analysis.4)Aiming at the problem that traditional telemetry data correlation analysis methods can only discover knowledge about the degree of correlation and can not provide relevant structural information,a cross-correlation structure knowledge discovery method based on mixed sampling,cost-sensitive matrix and eXtreme Gradient Boosting?XGBoost?is proposed.The experimental results show that the proposed method has higher classification accuracy in ROC?Receiver Operating Characteristic?curve and F1-score than back propagation neural network and long-term memory network,and is insensitive to unbalanced data.It is an effective method for knowledge discovery of cross-correlation structure of telemetry data.