Similarity Measure Of Time Series For Satellite Telemetry Data

Nowadays, there are numerous studies on satellite telemetry data to support automatic test data analysis, anomaly detection, fault diagnosis and prognostics in the field of aerospace engineering. Most of the studies are based on Euclide distance for the similarity measure of time series, in which the Euclide distance has many limitations on satellite telemetry data similarity measure, such as: it can not remove the effect of the high correlationship between different parameters, and it can not realize asynchronous measurement, etc. Aiming at solving these limitations, this thesis conducts research on time series similarity measures for satellite telemetry data.Firstly, aiming at the shortages when analyzing satellite telemetry data based on Euclide distance thar it can not remove the effect of the high correlationship between different parameters, and it can not realize asynchronous measurement, etc, we focus on time series similarity measure that can make up for the boundedness. A series of validation experiments are implemented on the open public datasets which have the similar characteristics with satellite telemetry data, to select the candidate similarity measures which are suitable for one-dimension and multi-dimention time series.Then, in order to verify the effectiveness of the candidate similarity measures for satellite telemetry data, we apply the s imilarity measures to anomaly detection method based on based on the hierarchical clustering and K-Nearest Neighbor classification. We realize the anomaly detection for satellite telemetry data and verify that the anomaly detection method based on Dynamic Time Warping distance can recognize anomaly time series with lesser difference at the same time.Finally, a data mining software platform is designed which can load data mining algorithms dynamically. The data mining algorithm library’s development and packaging are completed with standard interface and mixed programming technology of MATLAB and C language. As a result, the actual applications are realzied for time series similarity measures.The experimental results prove that applying more effective similarity measures to address satellite telemetry data can availably improve the performance of the anomaly detection. The developed algorithm library based on time series similarity measures has great potentials for actual applications. This work presents the better basis on both algorithms and applications for the future satellite telemetry data mining.