Predicting Airport Visibility using Seq2Seq Models

Accurate prediction of visibility at airports plays a crucial role in ensuring the safety of air transport. This paper establishes an airport visibility prediction model using the Seq2Seq network, which enables multi-step visibility prediction. The model's validity is tested and graded based on various indicators using Urumqi International Airport as a research site. The results indicate that the Seq2Seq model with attention mechanism surpasses MLP and other RNN models in terms of average absolute error, root mean square error, determination coefficient, and correlation coefficient score. Additionally, the Seq2Seq_GRU_A model, with its attention mechanism, is more effective in predicting low visibility conditions below 1 km, achieving the best prediction outcome with 1-hour TS and HSS scores reaching 0.54 and 0.7, respectively.