Attention-LSTM based prediction model for aircraft 4-D trajectory

Abstract Aviation activities are constantly increasing as a result of the growth of the global economic system. How to increase airspace capacity within the limited airspace resources while ensuring smooth and safe aircraft operations is a challenge for civil aviation today. Air traffic safety is supported by accurate trajectory prediction. The way-points are relatively sparse, and there are many uncertain factors in the flight, which greatly increases the difficulty of trajectory prediction. So, it is vital to enhance trajectory prediction accuracy. An attention-LSTM trajectory prediction model is proposed in this paper, which is split into two parts. The time-series features of the flight trajectory are extracted in the initial stage using the long-short-term memory neural network (LSTM). In the second part, the attention mechanism is employed to process the extracted sequence features. The impact of secondary elements is reduced while the influence of primary ones is increased according to the attention mechanism. We used the advanced models in trajectory prediction as the comparison models, such as LSTM, support vector machine (SVM), back propagation (BP) neural network, Hidden Markov Model (HMM), and convolutional long-term memory neural network (CNN-LSTM). The model we proposed is superior to the model above based on quantitative analysis and comparison.