Short-term air traffic flow forecasting based on model fusion

Short-term air traffic flow prediction provides decision information for optimal air traffic flow control and management. To accurately predict the short-term air traffic flow, this study uses time series decomposition to determine that the air traffic flow has obvious segmentation characteristics, that is, different time periods are superimposed with different degrees of periodicity, trend and randomness, where periodicity is mixed with two kinds of short-term and long-term circulation patterns. Existing prediction methods cannot capture the complex features of the traffic flow data dynamics well. Herein, we develop a new multi component network (MCNet) composed of a deep learning component and an autoregressive component. For capturing the periodicity of traffic flow and extract the short- and long-term recurrent patterns of traffic flow data, we use the deep learning component, consisting of a convolutional neural network and a recurrent neural network with a self-attention mechanism. The autoregressive component is responsible for catching the trend of traffic flow, solving the problem that the deep learning component is insensitive to the scale of input and output. Experiments are conducted on air traffic data based on OpenSky statistics, and the results show that MCNet achieves optimal results compared to other models.