Prediction of potential conflicts between active targets in airport flight zones based on LSTM

Abstract

Aiming at the problem of constant conflicts in airport flight zones, a method based on LSTM to predict potential conflicts of active targets in airport flight zones is proposed. Firstly, based on the complex network theory, two types of active targets, aircraft and vehicles, are taken as the research objects, and the flight zone active target network is established. The network dynamic evolution model is set up, and the operation data is input to calculate multiple network characteristic indicators. The principal component analysis of the indicator time series is performed to fit the potential conflict index; then, the Keras framework is used to build a long short-term memory neural network (LSTM) model, and the indicator time series is input into LSTM for training and prediction, and compared with other prediction methods; finally, the actual operation data of Xi'an Xianyang Airport is used for experiments, and the predicted values are compared with the true values. The mean square errors of the prediction results of each indicator are respectively 1.61%、13.13%、0.72%、0.004%、0.014%; The research shows that by establishing the flight zone active target network model, the network characteristic indicators can be used to characterize potential conflicts from different angles; LSTM can effectively predict the potential conflicts of the flight zone active target network, remind relevant personnel to pay attention to preventing conflicts and reducing the probability of conflicts.