基于机器学习的航空器进近飞行时间预测

为了准确预测航空器的落地时间，提高空管部门间的协作效率，采用机器学习的方法对航空器进近阶段飞行时间进行了预测。从实际运行出发，分析航空器在进近管制空域飞行时间产生差异的原因，提出了影响航空器在进近空域飞行的8类因素和17个重要特征。以航空器在进近飞行时间为标签，基于提出的重要特征，采用岭回归、支持向量机、随机森林和神经网络算法，建立了4种基于机器学习的航空器进近飞行时间预测模型。以南京进近为实例，对4种机器学习模型进行训练、验证和测试，对模型的性能指标、特征重要性和影响因素展开分析。研究结果表明，对于航空器进近飞行时间的预测，基于随机森林的模型表现出了最高的预测性能，模型的泛化能力最好、精确度高，回归效果越显著；进场状态是影响航空器进近飞行时间的最重要因素，而进场点和进场高度特征则对结果的贡献度最大。

Machine learning for aircraft approach time prediction

To improve the prediction accuracy of aircraft landing time, we propose a machine learning method to establish an approach time prediction model. Based on the actual operational situation, the primary reasons for different flying time in approach control are analyzed, including eight major factors with 17 important characteristics. Taking the aircraft flying time in approach airspace as labels, we use the proposed characteristics to build machine learning models with four popular machine learning algorithms:the ridge regression, random forest, support vector machine, and neural network. In the case of Nanjing Approach in China, four machine learning models are trained, validated and tested with practical operational data. The results show that the random forest based model exhibits the best prediction performance with good generalization ability, high accuracy and obvious regression effect. The initial arrival state of aircraft is the most important factor for approach time prediction, while the arrival point and initial altitude are two major characteristics for the prediction results.