To Catch Them All: A Generic Approach for Pattern Detection in Time Series Satellite Telemetry Data

Kurzfassung

This paper presents a tool that enables automatic recognition of patterns in historical satellite telemetry datasets. This tool uses pattern-matching techniques to identify substantially similar patterns to a user selected interval within a time series. In this sense, the software serves as a powerful data exploration tool for datasets which are too large for manual inspection. The system is designed to be robust to signal types, incomplete coverage, and inconsistent sampling, which are common issues with telemetry data. In detail, three distance-based algorithms, namely, Discrete Wavelet Transform, Dynamic Time Warping, and Adaptive Piecewise Constant Approximation, two probabilistic algorithms, namely, Gaussian Mixture Model, Hidden Markov Model, as well as an ensemble approach are implemented to cover and explore a wide variety of different pattern detection techniques. The system is evaluated on real satellite data and on the UCR (University of California, Riverside) Time Series Classification Archive.