基于ST-LightGBM的机场离港航班延误预测

Airport departure flight delay prediction based on ST-LightGBM

机场间的延误时空关系复杂,多数研究只聚焦于时间维相关性,导致延误预测精度不高.提出一种融合多机场时空相关性的ST-LightGBM模型预测机场离港航班延误.首先,构建多机场延误时空图数据;然后,通过图卷积神经网络提取延误信息空间特征,同时长短时记忆网络对机场各节点延误时间序列进行时间特征提取,形成具有时空相关性的二维特征向量;最后,将时空维特征向量输入LightGBM实现机场离港航班延误数量预测,在训练过程中引入贝叶斯优化算法进行参数寻优.结合真实数据实验,对中国枢纽机场延误数据进行时空维度关系提取并预测.结果表明,本文模型相比于其他基准模型具有较好的预测准确性.

The spatio-temporal relationship of delay between airports is complex,and most studies only focus on the correlation of time dimension,which leads to the low accuracy of delay prediction.This paper proposes a ST-LightGBM model that integrates the spatio-temporal correlation of multiple airports to predict airport departure flight delays.Firstly,the spatio-temporal graph data of multi-airport delay is constructed;Then,the spatial features of delay information are extracted by graph convolution neural network,and the temporal features of the delay time series of each node of the airport are extracted by long short term memory network to form a two-dimensional feature vector with spatio-temporal correlation;Finally,input the spatio-temporal feature vector into LightGBM to predict the number of airport departure flight delays,and introduce Bayesian optimization algorithm to optimize parameters in the training process.Combined with the real data experiment,the spatio-temporal dimensions of the delay data of the hub airports in China are extracted and predicted.The results show that this model has better prediction accuracy than other benchmark models.