A New Perspective on Airport Delay Prediction: A Three-Channel Temporal Convolution Network With Complex Network Information

Airport delay prediction plays a crucial role in air traffic management practices, including rerouting aircraft, implementing ground delays, and sequencing arrivals. This task is challenging due to the inherent nonlinear characteristics in traffic evolution. The popular deep learning-based traffic prediction methods lack the in-depth exploration of traffic evolution features. In this work, we propose a novel three-channel Temporal Convolution Network (TCN) framework incorporating temporal complex network feature information for airport delay prediction. The complex network feature sequence of the temporal networks can effectively capture the nonlinear dynamic behavior of airport delays. Firstly, we divide the original time series of airport delays into three distinct series (current, daily, and weekly) and feed them into three channels of our model. This operation allows the model to effectively capture the inherent characteristics of local proximity and global periodicity in airport delay time series. Secondly, the complex networks are converted from airport delay time series using complex network theory, and the topological features of networks are combined with three-chamel TCN, to improve the ability of learning short-term nonlinearity of airport delay evolution. Finally, we incorporate weather condition and flight schedule information as external features to further enhance the prediction accuracy. We perform extensive experiments on a flight dataset at Hartsfield-Jackson Atlanta International Airport (ATL), and the results demonstrate the superiority of our approach compared to existing benchmark methods.