航班撤轮挡里程碑时刻预测

发布时间：2018-03-12 16:46

本文选题：航班撤轮挡时刻预测　切入点：因子分析　出处：《中国民航大学》2017年硕士论文　论文类型：学位论文

【摘要】：航班撤轮挡时刻是所有舱门关闭,完成撤廊桥,推车可用,收到许可后立即推出的时刻,是航班进程监控的重要里程碑事件,能指导空管提前预判航班预起飞队列,以及机场和航空公司进行地面保障服务的重要时间节点。因此,航班撤轮挡时刻预测逐渐成为民航热点研究问题。现有预测方法采取经验统计方法得到航班撤轮挡时刻,其核心思想是首先统计历史数据得到平均最小过站时间,然后将预计降落时间、预计滑入时间和平均最小过站时间求和,以作为航班撤轮挡时刻的估计值。对于大型枢纽机场,不同的机型、航班时刻和地勤保障单位,航班过站时间有较大差异。因此,使用统一的过站时间预测航班撤轮挡时刻会产生较大误差,影响机场、航空公司和空管的协同运行效率。针对此问题,本文开展航班撤轮挡时刻预测研究,创新性采用机器学习方法构建航班撤轮挡时刻预测模型,而不再是历史数据的估计。直观上,可直接利用航班撤轮挡时刻的前序所有里程碑事件构建多元线性回归模型,但是不相关或弱关联的里程碑事件会对航班撤轮挡时刻预测产生偏差。针对此问题,本文提出了一个基于因子分析的航班撤轮挡时刻预测模型。首先,开展航班里程碑事件的因子分析,分析各里程碑事件间的相关性,通过计算因子载荷、因子旋转确定与航班撤轮挡时刻相关的关键里程碑事件。其次,将关键里程碑事件作为特征变量,利用多元线性回归的方法建立预测模型。实验表明,误差范围±5和±10分钟内,模型平均预测准确率分别可达70%和90%以上,高于A-CDM经验方法、基于主成分分析和基于支持向量回归(SVR)等航班撤轮挡时刻预测算法。由于基于因子分析的航班撤轮挡时刻预测模型本质上是融入1L正则化约束的多元线性回归问题。模型仅考虑前序里程碑事件对航班撤轮挡时刻的影响,忽略了航班登机口、天气、车辆调度及机场是否处于繁忙时刻等诸多难以量化因素的间接影响,无法更准确地预测航班撤轮挡时刻。针对上述情况,本文提出一个基于隐藏变量的航班撤轮挡时刻预测模型,将上述无法量化因素的影响通过隐藏变量在模型中体现。在训练阶段,由于最大化数据似然概率的优化目标耦合了模型参数,导致传统梯度上升等算法无法直接使用,为此采用变分EM算法求解模型参数,其中期望计算旨在优化求解近似分布的模型参数,而期望最大则是最大化似然概率求解回归预测模型参数。在基准数据集合上的实验表明,该模型比基于因子分析的航班撤轮挡时刻预测模型能够取得更好的均方误差以及准确率。
[Abstract]:The departure time is the time when all the doors are closed, the bridge is removed, the trolley is available, and immediately after receiving the permission, it is an important milestone in the monitoring of the flight process. It can guide the air traffic control to predict the flight pre-departure queue in advance. And the important time node for the ground support service of the airport and the airline. Therefore, the prediction of the departure time of the flight is gradually becoming a hot research issue in civil aviation. The existing forecasting methods adopt the empirical statistical method to get the time of the flight withdrawal. Its core idea is to first get the average minimum transit time by statistics of historical data, and then sum up the estimated landing time, the expected slide in time and the average minimum stop time to be used as the estimated value of the departure time of the flight. For a large hub airport, Different aircraft types, flight times and ground handling support units, and flight transit times are quite different. Therefore, the use of unified transit time to predict the departure time of a flight will result in greater errors, which will affect the airport. Aiming at this problem, this paper develops the research on the prediction of flight withdrawal time, and innovatively uses machine learning method to construct the model of flight withdrawal time prediction. It is not the estimation of historical data. Intuitively, multiple linear regression models can be built directly by using all the milestone events in the pre-order of the departure time of the flight. However, irrelevant or weakly related milestone events can cause deviation to the prediction of the departure time. In order to solve this problem, this paper proposes a prediction model based on factor analysis. Carries on the factor analysis of the flight milestone event, analyzes the correlation between each milestone event, through the calculation factor load, the factor rotation determines the key milestone event related to the flight withdrawal time. Secondly, Taking the key milestone events as characteristic variables, the prediction model is established by using multiple linear regression method. The experimental results show that the average prediction accuracy of the model can reach more than 70% and 90% within 卤5 and 卤10 minutes, respectively, which is higher than that of A-CDM empirical method. Based on principal component analysis (PCA) and support vector regression (SVR) algorithm, the prediction model of flight withdrawal time is essentially a multivariate linear regression model with 1L regularization constraint. The model only considers the influence of the pre-order milestone event on the departure time of the flight. Ignoring the indirect effects of many difficult factors, such as flight gate, weather, vehicle scheduling and whether the airport is in busy hours, it is impossible to predict more accurately the departure time of the flight. In this paper, we propose a prediction model of flight withdrawal time based on hidden variables. The influence of the above unquantifiable factors is reflected in the model by hidden variables. Because the optimization objective of maximizing the likelihood probability of the data coupled the model parameters, the traditional gradient rise algorithm can not be used directly, so the variational EM algorithm is used to solve the model parameters. Among them, the expected calculation is aimed at optimizing the model parameters for solving approximate distribution, while the maximum expectation is to maximize the likelihood probability to solve the regression prediction model parameters. The experiments on the set of datum data show that, This model can obtain better mean square error and accuracy than the prediction model based on factor analysis.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V355;O212.1

【参考文献】