Machine Learning-based Stability Assessment and Change Detection for Geosynchronous Satellites

摘要

Analysts have been able to manually inspect the light curve of a geosynchronous satellite to assess its stability - whether it is stable (three-axis-stabilized) or unstable (tumbling). However, with the large volume of data collected persistently with wide field of view sensors, manual inspection by humans may not be sustainable. It is desirable to automate the stability assessment to (a) classify the satellite as stable or tumbling and when possible (b) pinpoint the moment it transitions from stable to becoming unstable. In this paper, we will show how such an automated system is developed. We found the Random Forest (RF) of Decision Trees classifier to be sufficiently robust and accurate as a solution for (a) when the satellites are either stable or in an established (steady) state of tumbling, as evidenced by the high level of accuracy for stability assessment achieved by the RF. A discussion of the optimal features to use with RF is provided. Once the RF algorithm has detected the first night the satellite becomes unstable, we then aim to pinpoint the precise time of the change. Our trained RF alone is not sufficient for detecting the onset of tumbling because it requires and labels the entire light curve. Also with our existing data set, it was not sufficiently trained to recognize the state in the interim. During that time a combination of two tests is used to recognize a tumbling satellite. The periodicity test determines the significance of periodicity at the detected frequency. The normality test applied to residuals in the signature subintervals detects the presence of aliasing, which is caused by the fundamental tumbling frequency being higher than the observation sampling frequency. In the aftermath of the onset of instability, all three tests - RF, periodicity, and normality - are combined to update the satellite's status. We show the results of applying this set of algorithms on multi-year high cadence photometry of geosynchronous satellites.