Data-Driven Anomaly Detection for UAV Sensor Data Based on Deep Learning Prediction Model

Unmanned Aerial Vehicle (UAV) can accomplish various specific tasks and play an increasingly essential role in military, industrial and civil fields. However, the safety of the UAV is lower than that of manned aircraft, and great economic loss is caused due to its relatively high failure rate. Therefore, it is of great significance to study the anomaly detection method for the UAV system. In recent years, the deep learning method has been widely applied in various fields due to its outstanding advantages such as strong ability to approximate complex functions and automatic feature extraction. In this paper, a Long Short Term Memory (LSTM) Recurrent Neural Network method is proposed for the UAV anomaly detection. Firstly, a prediction model is formulated based on the training data set which contains only normal data, then the data at next time can be predicted. Secondly, according to the prediction results in train phase, we give an estimation of the prediction uncertainty. Finally, anomaly detection is achieved by comparing the prediction value with the uncertain interval. Real UAV sensor data with point anomalies in north velocity and pneumatic lifting velocity are used to verify the proposed method, and experimental results show that the proposed method can effectively detect point anomalies.