Flight Trajectory Prediction Based on LSTM

Abstract

Track prediction is a key technology to avoid potential navigation dangers and provide reliable scheduling. Due to the insufficient processing ability of traditional neural networks for long sequence data such as track data, the prediction error of aircraft trajectories is relatively large. In this paper, the Long Short-Term Memory network (LSTM) is used for 4D track prediction, and the excessive dependence between adjacent data is reduced through the sliding window method. The ADS-B track data is standardized, outliers are processed using the 3σ Principle, and the smoothness of the time series is ensured through cubic spline interpolation. Four-dimensional track prediction and two-dimensional prediction of each feature of Height, Speed, Speed, Longitude, and Latitude are carried out for the standard data. Performance is judged by Root Mean Squared Error (RMSE) and the average prediction error of each feature. Through comparison, the track prediction effect of the LSTM neural network is better than the existing BP neural network prediction method.