基于集成学习的航班延误等级预测方法  

Flight delay level prediction based on ensemble learning

作　　者：鲁亮 万欣然 樊玮[1] 郭威龙 LU Liang;WAN Xin-ran;FAN Wei;GUO Wei-long(College of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300

出　　处：《计算机工程与设计》2025年第4期1030-1037,共8页Computer Engineering and Design

基　　金：天津市教委科研计划基金项目(2021KJ046);中央高校基本科研业务费中国民航大学专项基金项目(3122019117)。

摘　　要：为提高航班延误预测的准确性,提出一种基于改进Stacking集成学习的航班延误等级预测方法。运用空中交通管理机场性能算法对出发和到达机场天气进行精确量化,引入机场相对繁忙度和前序航班等,对航班延误影响最为重要的因素构造特征,使用Catboost模型进行特征筛选,以及SMOTE与Tomek Link算法进行不平衡数据处理;在Stacking集成学习方法中引入基学习器权重参数进行建模,引入贝叶斯优化找到模型最佳的超参数组合。实验结果表明,改进后的方法相比原有方法在多项预测评价指标上均有提高。To enhance the prediction accuracy,a flight delay level prediction method based on improved Stacking ensemble lear-ning was proposed.The air traffic management airport performance(ATMAP)algorithm was utilized to accurately quantify the weather at departure and arrival airports.Factors crucial to flight delay,such as the relative busyness of airports and the impact of preceding flights,were considered to construct features.Catboost model was employed for feature selection,and synthetic minority over-sampling technique(SMOTE)combined with Tomek Link was applied for handling imbalanced data.In the Stacking ensemble learning method,the modeling was performed by introducing weight parameters for base learners,and Bayesian optimization was utilized to find the optimal combination of hyperparameters for the model.Experimental results demonstrate that the proposed improved method outperforms the original approach across multiple prediction evaluation metrics.

关 键 词：航班延误等级 预测模型 空中交通管理机场性能算法 数据不平衡处理 Stacking集成学习 权重参数 贝叶斯优化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]