A Multi-Head Self-Attention Transformer-Based Model for Traffic Situation Prediction in Terminal Areas

Abstract

Terminal operations management is an important part of air traffic management. Accurately detecting and predicting the operational status of the terminal area can help formulate more appropriate and efficient management methods. To achieve more accurate results in predicting the traffic situation, a ConvTrans-TCN (Convolutional Transformer with Temporal Convolutional Network) model is proposed in this paper. The model first constructs the feature extraction part using the causal-convolution multi-head self-attention module. It can effectively model the long-term dependency in the sequence and match the local patterns of the sequence, and it enhances the performance of feature extraction. Then the TCN (Temporal Convolutional Network) module is used to build the information fusion part to complete the fusion of feature data. The TCN architecture can accurately learn long-term and short-term dependencies in time series, and it has sufficient memory. Finally, the situation prediction is obtained by a feedforward neural network. The experiment’s results prove that this model is feasible and it performs better than the common models such as LSTM, BP, which can help air traffic managers to identify the operational status of the terminal area and provide decision support.

Keywords

• Intelligent transportation system
• air transportation
• traffic situation prediction
• transformer
• temporal convolutional network