A satellite telemetry data anomaly detection method integrating wavelet packet transform and diffusion models (WPT-diffusion)

In addressing the challenge of limited anomaly detection methods for high-dimensional, non-stationary satellite telemetry data, researchers have developed a novel approach called wavelet packet transform–diffusion (WPT-diffusion). This method integrates multi-scale time-frequency features as structural conditional guidance within a diffusion generative model.

The researchers introduce a learnable Morlet wavelet kernel and frequency-domain attention mechanism to boost the sensitivity of the model to subtle spectral anomalies. Additionally, a temporal attention mechanism is employed to capture long-range dependencies, thereby enhancing reconstruction fidelity.

Testing this approach on the ESA OPS-SAT real-world dataset, the WPT-diffusion method demonstrated impressive performance, achieving a precision of 0.90, a recall of 0.88, and an F1-score of 0.890. These results surpassed those of other existing anomaly detection methods such as CATCH, USAD, TimesNet, TranAD, and DDTAD. The effectiveness of the WPT-diffusion method was confirmed through these promising results, showcasing its potential for application in real-world satellite mission scenarios.

This research was supported by the National Natural Science Foundation of China, with Grant Number 62472437. The study was conducted by a team from the School of Information and Navigation, Air Force Engineering University, and the National Key Laboratory of Unmanned Aerial Vehicle Technology in Xi'an, China. The article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which allows non-commercial use and sharing of the material as long as proper credit is given to the original authors and source.