【摘要】 安全是民航业的生命线,也是民航永恒的主题。本文面向终端区航班流,以空中交通复杂性及航空器潜在冲突关系为切入点,研究终端区航空器运行安全态势,提出了一种基于深度学习的终端区航班流安全态势感知方法。首先,提出更为全面和准确的安全态势评估特征;其次,构建一种添加安全态势信息捕捉层的深度聚类态势识别模型;最后,基于注意力机制构建时空图卷积神经网络的安全态势等级预测模型。通过真实数据集实验结果对本文所提方法进行评估,发现:(1)本文所提模型在各方面性能上优于传统模型;(2)所提态势识别模型能够确保编码特征可以捕捉到安全态势的区分性信息,增强模型的可解释性与识别任务的匹配性;(3)所提态势预测模型具有更优秀的空间和时间的综合建模能力。最后本文揭示了空中交通安全态势的时空演变特性,为空中交通安全管理提供参考。更多还原

【Abstract】 Safety is the cornerstone of the civil aviation industry and the enduring focus of civil aviation. This paper uses air traffic complexity and potential aircraft conflict relationships as entry points to study the operational safety level of terminal area flight flows and proposes a deep learning-based method for safety situation awareness in terminal area aircraft operations. Firstly, a more comprehensive and precise safety situation assessment features are constructed. Secondly, a deep clustering situation recognition model with added safety situation information capture layer is proposed. Finally, a spatiotemporal graph convolutional neural network based on attention mechanism is constructed for predicting safety situations. Experimental results from a real dataset show that:(1) The proposed model surpasses traditional models across all evaluated dimensions;(2) the recognition model ensures that the encoded features capture distinctive safety situation information, thereby enhancing model interpretability and task alignment;(3) the prediction model demonstrates superior integrated modeling capabilities in both spatial and temporal dimensions. Ultimately, this paper elucidates the spatiotemporal evolution characteristics of air traffic safety situation levels, offering valuable insights for air traffic safety management.更多还原