STL combining LSTM for long-term predicting airport traffic flow

Airport traffic flow exhibits significant periodicity on a daily scale, few studies have given attention to periodicity when predicting airport traffic flow. In this article, we propose a novel model that combines long short-term memory (LSTM) and seasonal-trend decomposition procedure based on loess (STL) to predict the arrival/departure traffic flow at the airport. A sinusoidal template-matching method based on Fréchet distance is used to restack the periodic input variables. A time series decomposition algorithm STL is used to decompose the traffic flow time series into trend, seasonal, and remainder components to identify its periodic structure. LSTM model is trained using historical airport operation data, strategic flight schedule data, and meteorological data from Beijing Capital International Airport, Guangzhou Baiyun International Airport, and Shanghai Pudong International Airport in 2019. Our results demonstrate that both the adaptive restack of input variables and the time series decomposition algorithm STL can improve the prediction performance. Our proposed method shows superior performance in long-time prediction (720-time steps). In particular, STL combined LSTM method achieves an R-squared of 0.97 and a mean absolute error (MAE) of less than 1.58 for all three airports.