Real-time Short- Term Trajectory Prediction Based on GRU Neural Network

Abstract

From the perspective of historical trajectory data and real-time trajectory data, a short-term real-time trajectory coordinate point prediction method based on GRU (Gated Recurrent Unit) cyclic neural network is proposed. The main method of real-time trajectory prediction algorithm, in the first stage, is neural network initialization training. The GRU neural network parameters are learned by batch processing, and the network parameters which have been trained are input to the second stage as the initial parameters of the next stage GRU neural network. The second stage is the real-time prediction of the trajectory. The GRU neural network parameters trained in the first stage are used as the initial values of parameters of the online prediction network. The real-time trajectory data is used to adaptively update the parameters of the GRU neural network online and real-time predicted the flight points. Finally, compared with other typical time series prediction models based on real flight data simulation experiments, the method is verified that the proposed short-term real-time prediction model has obvious advantages in prediction accuracy and applicability.