基于集成学习的航空器滑出时间预测研究  

Research on Prediction of Taxi-Out Time of Aircraft Based on Ensemble Learning Methods

作　　者：白晓妮 宫献鑫 阮妨 延梦璐 BAI Xiao-ni;GONG Xian-xin;RUAN Fang;YAN Meng-lu(Northwest Civil Aviation Design Institute Project Management Co.Ltd,Xi'an Shaanxi 710075,China;Office of Aviation Safety Civil Aviation Flight University of China,Guanghan Sichuan 510903,China)

机构地区：[1]西安西北民航项目管理有限公司,陕西西安710075 [2]中国民用航空飞行学院航空安全办公室,四川广汉510903

出　　处：《计算机仿真》2024年第8期13-18,181,共7页Computer Simulation

摘　　要：为提升航空器离场滑出时间的预测精度,分析了影响滑出时间的各类因素,引入场面运行状况和气象条件两类特征,基于装袋方法、随机森林、自适应增强和梯度提升等四种典型的集成学习方法,构建了滑行时间预测模型。以美国肯尼迪机场为算例,采用判决系数、RMSE和MAE等性能度量指标验证算法预测性能。实验结果表明:气象特征的引入能够提升滑出时间预测精度;与其它回归算法对比,集成学习的预测误差较小;分析集成方法下的学习曲线发现自适应增强和梯度提升方法能够有效避免过拟合现象。研究结果可用于集成化场面管理软件的开发应用。In order to improve the prediction performance of taxi-out time,various factors impacting taxi-out time were analyzed,and then two kinds of features(surface operating conditions and meteorological conditions)were introduced into our taxi-out prediction models,which were built based on ensemble learning algorithms including bagging method,random forest,Adaptive Boosting and Gradient Boost Machine.Taking JFK as an example,performance metrics such as coefficient of determination,RMSE,and MAE were used to verify the prediction performance of the algorithms.The experimental results show that the introduction of meteorological features can improve the prediction accuracy of taxi-out time;the prediction errors of ensemble learning are smaller than other regression algorithms;the learning curve under the ensemble methods are analyzed and we find that AdaBoost and GBM can effectively avoid overfitting.The research results can be used in the development and application of integrated surface management software.

关 键 词：空中交通流量管理 滑出时间 预测性能 集成学习 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]