Air Traffic Flow Forecasting Based on Spatial-temporal Graph Network
Accurateair traffic flow forecasting is significant for air transport management and flight safety.However,precise forecasting of air traffic flow is challenging due to complex temporal fluctuation patter-nand close dependency between different airports.This paper proposedan air traffic flow forecasting ap-proach based Spatial-Temporal Graph Network to capture both temporal variation pattern and dependency between different airports.In the spatial feature learning module,graph ordinary differential equation was employed to extract the dependency relationships between airports.In the temporal feature learning mod-ule,the efficient reformer was introduced to characterize the long-range temporal correlation of air traffic flow.In the dataset of Airline on-the-time Performance Data,the proposed method achieved better per-formance than the compared forecasting methods.The proposed method can achieve the performance with the WMAPE(Weighted Mean Absolute Percentage Error)metrics of 35.51%,36.54%and 35.55%,for prediction at the length of 6 h,9 h and 12 h.
air traffic forecastingtimeseries forecasting,graph representationTransformer methodspatial-temporal dependencies
万方数据
维普
