摘要
针对航班地面保障数据缺失问题,提出一种基于图神经网络的数据插补算法。通过降噪编码器降低原始数据中噪声对训练的影响,增强提取特征的可靠性;建立一种图表示学习框架,使用聚合函数聚合采样区间内节点的特征,实现神经网络节点的状态更新,得到第1次嵌入特征;应用长短时记忆网络对航班的时序信息进行第2次嵌入得到隐藏层的状态空间;通过反卷积神经网络进行特征还原,提出一种损失函数实现网络的迭代,在迭代多次后得到最终的航班地面保障数据插补结果。使用西南某机场2018年4~6月份的航班地面保障数据对所提算法进行测试,结果表明:相比于其他算法,所提算法在低缺失率时,插补误差平均降低了约74%;在较高缺失率时,插补误差平均降低了约68%;所提算法迭代次数约在100次,正则化系数约为0.5时,插补误差达到最低。
A data imputation algorithm based on a graph neural network is proposed to address the issue of missing flight ground support data.Firstly,to reduce the impact of noise in the original data on training denoising autoencoder is applied to enhance the reliability of feature extraction.Secondly,a graph representation learning framework is established to get the first embedding,using aggregation functions to aggregate the features of nodes within the sampling interval to achieve state updating.Furthermore,a long and short-term memory neural network is constructed to embed the temporal feature of flights to obtain the final state space of the hidden layer.Lastly,a loss function is suggested to iterate the deconvolution neural network,which is employed for feature restoration.The final flight ground operation data imputation result was acquired after numerous iterations,and the technique was evaluated using ground operation data from a specific airport in Southwest China from April to June 2018.The results showed that compared to other algorithms,the proposed algorithm imputation error decreased by an average of about 74%at low missing rates.At higher missing rates,the imputation proposed algorithm error decreased by an average of about 68%.When the number of iterations of the proposed algorithm is about 100 and the regularization coefficient is about 0.5,the imputation error reaches the lowest.
作者
邢志伟
孙恪
罗谦
刘畅
张涛
乔迪
XING Zhiwei;SUN Ke;LUO Qian;LIU Chang;ZHANG Tao;QIAO Di(Electronic Information and Automation Institute,Civil Aviation University of China,Tianjin 300300,China;Engineering Technology Research Center,The Second Institute of Civil Aviation Administration of China,Chengdu 610041,China;Chengdu Tianfu International Airport Branch,Sichuan Airport Group Co.,Ltd.,Chengdu 610041,China)
出处
《北京航空航天大学学报》
北大核心
2025年第5期1528-1538,共11页
Journal of Beijing University of Aeronautics and Astronautics
基金
国家重点研发计划项目(2018YFB1601200)
国家自然科学基金青年科学基金(62203452)
中国民航大学研究生科研创新资助项目(2022YJS105)。
关键词
航班地面保障
图嵌入
缺失值插补
递归神经网络
自动编码器
多元时间序列
flight ground operation
graph embedding
missing value imputation
recurrent neural network
autocoder
multivariate time series
