Anomaly Detection of Spacecraft Telemetry Data Using Temporal Convolution Network

Abstract

Spacecraft telemetry data is the only basis for the ground management and operation system to determine its on-orbit status. The rapid development of on-orbit spacecraft generates a large amount of telemetry data with high dimensionality and complex correlation between variables, which brings great challenges for current and future ground management and operation for these spacecrafts. As a result, its anomaly detection becomes an essential solution for enhancing the operation safely and reliably and realizing intelligent maintenance. The above issues make it a hot spot research topic for realization of high detection rate, low false detection rate, and strong interpretability of anomaly detection of telemetry data. In further, timely detection of abnormalities can reduce spacecraft unexpected accidents. Recently, Temporal Convolution Network (TCN) has arisen much attention for telemetry time series data modeling by its good parallel processing ability and temporal characterization. Thus, this paper proposes an anomaly detection model for spacecraft telemetry data based on TCN time series prediction. Firstly, TCN algorithm is used to extract features of telemetry data and make a prediction. Then, the potential anomalies are determined by the static threshold judgement method. Actual satellite telemetry data are used to implement performance evaluation and comparison. The proposed method is compared with Long Short Term Memory (LSTM) algorithm which is also a widely utilized algorithm in recent time series analysis. Four metrics are adopted for the experiments: the detection accuracy, detection rate, false detection rate and model inference time. TCN achieves high accuracy, high detection rate and low false detection rate. In addition, TCN save much more model inference time than LSTM with long timestep. Generally, TCN is superior to LSTM on operating efficiency.