回帰学習による逐次予測を用いた宇宙機の適応的リミットチェック

Abstract

This paper proposes a novel “knowledge-free” anomaly detection method for spacecraft based on regression techniques. This method learns a linear or nonlinear probabilistic regression model in the learning phase by applying a regression technique to a massive telemetry data of spacecraft, and then monitors the real-time telemetry data using the constructed model. This approach can be seen as adaptive limit-checking because it sequentially predicts proper envelop of a target time-series based on the past data of itself and other relevant series. We have confirmed the effectiveness of the proposed anomaly detection method by applying it to the telemetry data obtained from a simulator of an orbital transfer vehicle designed to make a rendezvous maneuver and from actual space operation of Engineering Test Satellite VII (ETS-VII).