关键词: 航空运输, 离港聚集客流预测, 完全自适应噪声集合经验模态分解, 长短期记忆神经网络, 航站楼客流, 航班延误特征

Abstract: To improve the accuracy and efficiency of predicting passenger flow for resource planning and passenger management within terminal buildings during flight delays, this paper proposes a departure aggregation passenger flow prediction method considering flight delay characteristics. The method introduces the flight delay parameters to quantitatively characterize the fluctuation of the departure aggregation passenger flow in the airport terminal (DAPFT). The fluctuation pattern and distribution characteristic of departing aggregation passenger flow are analyzed under flight delay. A short-term terminal aggregation passenger flow prediction model is proposed based on adaptive noise complete ensemble empirical modal decomposition (CEEMDAN), permutation entropy algorithm (PE), and whale optimization algorithm (WOA) optimised long- short- term memory neural network (LSTM). The CEEMDAN is applied to decompose the aggregation passenger flow data series into several modal components intrinsic mode function (IMF) and a residual Res to reduce the complexity and non-stationarity of the data of the original series. To reduce the computational scale of the model and improve the prediction efficiency and accuracy at the same time, the PE algorithm is used to calculate the entropy value of each IMF component and reconstruct the components based on the entropy value. Then, the WOA-LSTM (W-L) passenger flow aggregation prediction model is established, the whale optimization algorithm is used to optimize the LSTM's hyperparameters, and the reconstructed components' predictions are superimposed to obtain the final aggregation passenger flow prediction target value. The model has been applied to a hub airport in the Yangtze River Delta for predictive performance validation. The results show that the CEEMDANPE-WOA-LSTM (C-P-W-L) prediction model has the best performance, and compared with the simple LSTM model, the root mean square error is reduced by 42.78%, the average absolute error is reduced by 44.00%, and the percentage error of the prediction of departure hall aggregation passenger flow(DHAPF), is reduced by 45.62%. The prediction efficiency is improved by 41.64% compared with the CEEMDAN-WOA-LSTM (C-W-L) model. The proposed model can effectively fit the departure hall aggregation passenger flow data with significant nonlinear and non- stationary characteristics, and has high prediction accuracy and computational efficiency.

Key words: air transportation, forecasts of departing aggregation passenger aggregation flow, complementary ensemble empirical mode decomposition with adaptive noise, long short-term memory neural network, terminal passenger flow; flight delay characteristics