Predicting pilot behavior during midair encounters using long short-term memory network

Abstract

Characterized by the wide use of advanced automation and the introduction of new operation concepts, the future air transportation system will be more complex. Advanced pilot behavior models with improved capability are required to support the design and analysis of the midair encounter situations in the future air transportation system. This paper first filters midair encounter data from Automatic Dependent Surveillance-Broadcast (ADS-B) observations. Based on the acquired midair encounter data, a comprehensive pilot behavior model is proposed based on a multi-layer Long Short-Term Memory (LSTM) network. The model is designed for the purpose of enhancing the predicting capability of pilot behaviors in both horizontal and vertical planes. Finally, the performance of the proposed model to predict pilot behavior in both horizontal and vertical planes is studied through evaluating against realistic midair encounter situations.