基于DBN-SVM的航班延误内在模式分析  

Internal Pattern Analysis of Flight Delay Based on DBN-SVM

作　　者：朱代武[1] 陈泽晖 刘豪 ZHU Dai-wu;CHEN Ze-hui;LIU Hao(Civil Aviation Flight University of China,Guanghan 618000,China)

机构地区：[1]中国民用航空飞行学院,四川广汉618000

出　　处：《航空计算技术》2022年第1期36-40,共5页Aeronautical Computing Technique

基　　金：民航局安全能力建设项目项目资助(14002600100015J013)。

摘　　要：空中交通系统作为典型复杂系统,其非线性聚合的动力学特征给延误预测带来挑战,使延误预测问题保持着开放性。对于航班延误这类考虑多因素的问题,研究采用深度置信网络支持向量机(DBN SVM)回归方法建立航班延误预测模型,方法用来挖掘航班延误的内在模式,将支持向量回归嵌入到开发的模型中,使其能够在提出的预测体系结构中执行有监督的微调,并将交通管理措施(TMI)中一些关键影响因素,作为高斯伯努利(GBRBM)的隐藏层,作为模型的下一个可见层,将TMI关键因素添加至模型中,有助于减少整体延迟。对于测试集的不平衡高维数据集,研究将采用准确性,敏感性来评估因变量和解释变量之间的关系,最后数据表明DBN SVM模型的延误预测准确度达到89.39%,可为流量管理自动化计算提供一定理论依据。As a typical complex system,the dynamic characteristics of nonlinear convergence of air traffic system challenge delay prediction and keep the problem of delay prediction open.To consider many factors,such as flight delay problems,this study adopts the deep belief networks support vector machine(DBN SVM)regression method to establish flight delay prediction model,the method is used to mining delays intrinsic mode,to embed the support vector regression model of development,make its can be estimated in the proposed architecture perform supervised fine tuning,In addition,some key influencing factors in the traffic management measures(TMI)are added into the model as the hidden layer of Gauss Bernoulli(GBRBM)and the next visible layer of the model,which helps to reduce the overall delay.For the unbalanced high dimensional data set of the test set,this study will use accuracy and sensitivity to evaluate the relationship between dependent variables and explanatory variables.The final data show that the accuracy of delay prediction of DBN SVM model reaches 89.39%,which can provide a theoretical basis for automatic calculation of flow management.

关 键 词：空中交通管理 航班延误预测 深度置信网络支持向量机 数据挖掘 

分 类 号：V355[航空宇航科学与技术—人机与环境工程]