Flight Delay Propagation Prediction Based on Deep Learning

The current flight delay not only affects the normal operation of flight, but also spreads to downstream flights through schedule, resulting in a wide range delays. analysis and prediction propagation advance can help civil aviation departments control rate reduce economic loss caused by Due small number data samples that constitute chains, it is difficult construct chain data. In recent years, problem generally based on traditional machine learning methods with sample size. After obtaining large amount raw from China Air Traffic Management Bureau, we have constructed 36,287 pieces three-level Based these data, tried use deep method analyze forecast field learning, there are CNN models RNN deal classification problems well. two classes models, modify innovate study prediction. Firstly, CNN-based CondenseNet algorithm used predict level this, network improved inserting CBAM modules named CBAM-CondenseNet. experimental results show effectively improve performance, accuracy reach 89.8%. Compared method, average increased 8.7 percentage points. On basis model, considered superiority LSTM (Long Short-Term Memory network) considering processing time sequence information, then CNN-MLSTM injected SimAM module enhance attention experiment prediction, 91.36%, which significant improvement compared using or alone.