In view of the increasing number of unmanned aircraft vehicles (UAVs) in low altitude airspace, which leads to the increasing difficulty of surveillance, a real-time trajectory prediction method based on the gated recurrent unit (GRU) is proposed. The parameters of the GRU model are determined including the number of hidden layers, the cell sizes of a single hidden layer and the time step. Historical trajectory data are used to train and test the model. The result shows that the mean absolute errors (MAE) of altitude, longitude, and latitude predicted by the GRU model are 0.86 m, 2.85 m, and 3.65 m, respectively. The prediction results of the GRU model are compared with the long and short-term memory (LSTM), the support vector regression (SVR) and the autoregressive integrated moving average model (ARIMA). It is found that the GRU and LSTM outperform the two other models. The LSTM prediction accuracy is slightly higher than that of the GRU model. However, the space complexity of the LSTM model is much larger than that of the GRU model, and its training time is much longer.
Real-time Trajectory Prediction of Unmanned Aircraft Vehicles Based on Gated Recurrent Unit
Lect. Notes Electrical Eng.
14.12.2021
12 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Real-time Trajectory Prediction of Unmanned Aircraft Vehicles Based on Gated Recurrent Unit
British Library Conference Proceedings | 2022
|