A Short-term Prediction of Air Traffic Flow Based on a Wavelet-optimized GRU-ARMA Model

A short-term prediction of air traffic flow is important for air traffic management, and effectively reduce traffic congestion.To improve the accuracy of the short-term prediction and reduce the workload of air traffic controllers, a wavelet-optimized GRU-ARMA based model is proposed.Based on traditional prediction methods, the originaldata of air traffic flow is decomposed by multi-scale wavelet transform. The detailed features of traffic flow with different frequenciesare extracted. Moreover, by using wavelet transform, component at low frequencies is subdivided as trend term, and time at high frequencies as noise term.Among them, the trend term represents the overall evolution trends of air traffic flow over time, while the noise term describes the comprehensive influences of random factors on air traffic flow. The gated recurrent unit (GRU) neural network and the autoregressive moving average (ARMA) model are used to predict the trend and noise terms, respectively.The prediction values of trend and noise terms are superimposed to obtain the final value of short-termprediction. An error analysis shows that the method maintains a stable prediction of about 2% at each prediction point. In contrast, the models that directly use raw traffic data for prediction (i.e. GRU, BiLSTM, CNN-LSTM neural network models) and the single ARMA model have prediction errors ranging from 5% to 37.14%.Compared to the GRU, BiLSTM and CNN-LSTM neural network models, the prediction accuracy of the proposed model is increased by 3.02%, 5.39% and 5.05%, respectively.