Aircraft Trajectory Prediction Based on Residual Recurrent Neural Networks

Abstract

Accurate and efficient aircraft trajectory prediction is one of the important technologies developed in the aerospace field today. The high space-time complexity, strong uncertainty and changeable flight trajectories of aircraft flight have brought great difficulties to the modeling and solution of intelligent trajectory planning. Aiming at the problems of strong maneuverability and difficult trajectory prediction of unpowered aircraft, based on the analysis of trajectory characteristics, this paper proposed a trajectory prediction method for aircraft based on residual recurrent neural network (RESRNN) for trajectory prediction. First, our algorithm decomposes the 3D trajectory data of the aircraft in time series. Then, this paper uses RNN for loop calculation and RESNET for residual extraction to get the predicted result. Furthermore, in order to solve the problem of lack of simulation sample libraries, this paper proposes a sample generation method based on differential dynamics to generate a sample library for validating our algorithm. The simulation results show that compared with other prediction methods, our method has higher prediction accuracy, which has certain reference significance for intelligent trajectory planning, trajectory prediction, and interception of other large maneuvering targets.