Spacecraft Telemetry Data Anomaly Detection Based On Multi-objective Optimization Interval Prediction

Abstract

Spacecraft telemetry data anomaly detection is crucial for the timely detection of potential malfunction in spacecraft systems. Because of the uncertainty of prediction, interval prediction models are more suitable for anomaly detection than point prediction and probability prediction. This paper first puts forward an anomaly detection framework based on the traditional LUBE model, and introduces a method to eliminate the error of the model itself in the framework of anomaly detection. Considering that the LUBE method judges the quality of the prediction interval, there are two indicators, interval width and interval coverage, which is essentially a multiobjective optimization problem. Therefore, this paper proposes a LUBE interval prediction model based on multi-objective optimization. Compared with the traditional model, the combination of the two indicators is obviously superior to the original method. Finally, the effectiveness is proved by anomaly detection experiments of public datasets and spacecraft telemetry data.