Flight Trajectory Prediction Model Based on DI-STGCN and Bayesian Safety Assessment Method

With the increasing flight flow in terminal areas of airports, accurate prediction of aircraft tracks is one of the effective methods to improve airport safety management. A trajectory prediction model based on dynamic identification spatial-temporal graph convolution neural network (DI-STGCN) is used to process the data acquired by the broadcast automatic dependent surveillance system (ADS-B), and train and predict it. The results show that the predicted errors of the DI-STGCN model in longitude, latitude and altitude are slightly better than those of the LSTM model, and significantly better than those of the SVR and XGBoost models. Thus, the trajectory prediction based on the DI-STGCN algorithm has a high accuracy. Finally, the real-time safety of each flight stage is evaluated effectively by combining the Bayesian neural network and entropy weight method.