Anomaly Detection and Trajectory Prediction Methods for Substation UAV

Due to the impact of the complex environment of substations, the flight safety of UAVs is an increasingly important issue when performing inspection tasks at substations. Currently, the flight status monitoring of UAVs mainly uses the flight control system, which lacks in-depth analysis of flight data, especially flight trajectories data. In order to dig deeper into the flight control system data, anomaly detection and trajectory prediction methods are proposed in this paper. Firstly, based on the subspace learning method, we learn the local timing information of the original data saved in the subspace, which not only reduces the computational complexity of the data but also facilitates the discovery of anomalous data. Secondly, the anomaly detection of flight data is achieved by measuring the changes of data subspace vectors. Finally, UAV trajectory prediction is achieved by LSTM neural network based on time series, which only warns the anomalous data that will cause UAV flight safety and improves the warning efficiency.