An encoder-decoder generative adversarial network-based anomaly detection approach for satellite telemetry data

Abstract

Telemetry anomaly detection is a prominent health condition monitoring task that plays an increasingly crucial role in discovering potential incidents and facilitating long-term reliable on-orbit operations of satellite. Nevertheless, multiple monitored telemetry parameters and the complicated correlations among them pose significant challenges. In this study, an encoder-decoder generative adversarial network (EDGAN)-based anomaly detector is proposed for detecting multivariate anomalies in telemetry data. Firstly, an additional encoder is merged into the generator to learn a latent representation of the multivariate telemetry data for generating a reasonable and enhanced substitute of the data. Secondly, a feature matching strategy is applied in the discriminator to facilitate extracting more distinguishable features. Thirdly, three loss functions are designed for model training, and the deviation from the data space of the generator is utilized as the criterion to detect anomalies. Finally, experiments on real-world and public satellite datasets are conducted and the results verify the effectiveness of the proposed method.