卫星遥测数据相关性知识发现方法

Correlation knowledge discovery method for satellite telemetry data

为快速发现海量遥测数据中的相关关系,提出一种基于改进最大信息系数(Maximal Information Coefficient,MIC)的遥测数据相关性知识发现方法。以Mini Batch K-Means聚类算法为前驱过程对数据进行网格划分;计算该网格划分下的互信息,并以信息熵代替原有最大熵对互信息进行归一化矫正得到信息系数;选择不同网格划分下MIC作为变量相关性的测度。采用量子卫星遥测数据进行试验,结果表明:与基于动态规划算法的MIC方法相比,所提方法可有效解决MIC测度偏向多值变量的问题,时间复杂度从O(n^2.4)下降为O(n^1.6),是一种适用于大规模遥测数据相关性分析的有效方法。

To discover correlations in massive telemetry data efficiently,a novel correlation knowledge discovery method based on the improved MIC(maximal information coefficient)was proposed.The Mini Batch K-Means clustering algorithm was used to discretize data in the precursor process;the mutual information between two variables under this partition was calculated and normalized by information entropy instead of maximal entropy to obtain the information coefficient;the MIC was selected as the measure of variable correlation.Aflerwards,the method was applied to the correlation analysis of the quantum satellite telemetry data,and the results show that the 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 O(n^2.4)to O(n^1.6),and it is an effective method for large-scale telemetry data correlation analysis.