An Effective Method for Mining Quantitative Association Rules with Clustering Partition in Satellite Telemetry Data
Xin Dong, Dechang Pi · 2014
The correlation analysis of telemetry data plays a significant role in satellite performance analysis. However, the existing methods cannot be well applied, because the telemetry data is large and high-dimensional. In this paper, an efficient algorithm named QARC Apriori is proposed. First, to reduce the redundant attributes and lower the problem complexity, grey relational analysis method is applied. Second, each filtered attribute is partitioned into several subintervals, combining with K-Means clustering algorithm. During clustering, the outliers are removed to improve the accuracy of clustering results. Due to different distributions and scopes of attributes, the clustering centers are automatically adjusted. Moreover, the statistical information of each attribute is used to avoid repeatedly scanning database. Finally, all quantitative association rules are mined by an improved Apriori algorithm. In order to improve the mining efficiency, two pruning strategies are used. The experiments are conducted with the power supply data of a China's satellite from 2011.6.1 to 2011.9.1. It indicates that the proposed algorithm is suitable for quantitative association rules mining and is important for satellite on-orbit performance analysis.