Satellite Telemetry Data Anomaly Detection Using Bi-LSTM Prediction Based Model

Abstract

Satellite telemetry data is regularly received by the ground station, and then the ground staff determines the potential operating risk by checking real-time telemetry data. Among these data analysis tasks, anomaly detection is necessary to determine potential failure risk or early fault. Nowadays, the ground operator usually used the historical experience to set a fixed alarm threshold to judge the telemetry data. However, lots of complex abnormal data are difficult to capture in time and accurately by setting experiential alarm threshold. Thus, in this work, a telemetry time series data anomaly detection method is proposed based on bi-directional long short-term memory neural network (Bi-LSTM). The proposed method applies the strong temporal feature extraction capability to model and regress the satellite data. As a result, time series prediction using this model can implement the point data anomaly detection by evaluating the predictor and real value. To improve the suitability of the prediction based model, a dynamic threshold optimization method is also focused and integrated into the proposed framework. Experimental results with three satellite telemetry data sets prove the effectiveness of the proposed method. In addition, the satisfied performance is verified by comparing with RNN and basic LSTM model.