Terminal Area Traffic Situation Prediction using a Transformer-based Model with Multi-Head Self- Attention

Terminal areas such as airports, ports, and logistics centers often face complex traffic conditions due to continuous vehicle flow and unpredictable patterns. Addressing these issues requires accurate and timely forecasting to reduce congestion, enhance safety, and ensure efficient resource utilization. This study introduces a forecasting model based on the Multi- Head Attention Transformer architecture, which leverages self-attention mechanisms to identify complex temporal relationships in traffic data. By using multiple attention heads, the model captures patterns across different time scales, improving the prediction of both shortterm variations and long-term trends. When evaluated against traditional models like ARIMA, LSTM, and GRU, the proposed approach demonstrated notably higher accuracy. While the model shows strong potential for real-time traffic management in various terminal settings, it also demands extensive data and computational power to function optimally.