基于机器学习的航班静态四维轨迹预测研究  

Machine learning based algorithm for static 4D trajectory prediction of flight

作　　者：梁海军[1] 韩琪聪 LIANG Haijun;HAN Qicong(College of Air Traffic Management, Civil Aviation Flight University of China, Guanghan 618307, China)

机构地区：[1]中国民用航空飞行学院空中交通管理学院,四川广汉618307

出　　处：《兵器装备工程学报》2022年第4期147-151,共5页Journal of Ordnance Equipment Engineering

基　　金：四川省科技计划资助项目(2022YFG0210);国家自然科学基金委员会与中国民用航空局联合资助项目(U1433126)。

摘　　要：提出一种基于机器学习的方法预测航班飞行前的静态四维轨迹,较已有的运动学和动力学预测方法结果精确、稳定。该方法在真实历史监视数据的基础上运用隐马尔科夫模型对航空器飞行过程中位置和高度状态转移建模。在经纬度平面上以网格化地图为观测值、关键点航段位置为隐状态建模;在高度方向上以固定高度间隔为观测值、标准飞行高度层为隐状态建模。在运用EM算法学习到预测模型最优参数之后对航班经过各关键点的时间和高度进行预测,为航班管理提供有力的辅助支撑。通过数据仿真实验表明,采用本文提出的算法计算得到的结果较传统的运动学和动力学预测方法具有更高的精度和稳定性。The main purpose of 4D trajectory prediction of flight plan is to estimate the fly over time and altitude at every key point of route.The existing methods struggle with the prediction error and stability.A machine learning based algorithm was presented in this paper to predict the static 4D trajectory prediction for certain flight.The algorithm models the position and altitude transition by Hidden Markov Model(HMM)during flying on the basis of collected historical data.In the horizontal plate,the gridded map and route segment between key points were regarded as observation and hidden states.And the altitude separation and standard civil aviation fly level were processed as observation and hidden states.After optimizing the model parameters by Expectation Maximum(EM),the 4D trajectory was predicted by the learned model.Based on the trajectory,the relevant departments can take more efficient measurement for flight management.The simulation result with real collected data shows that the proposed algorithm poses higher accuracy than the traditional ones with more stable.

关 键 词：四维轨迹预测 机器学习 历史监视数据 隐马尔科夫模型 

分 类 号：V355[航空宇航科学与技术—人机与环境工程] TP391[自动化与计算机技术—计算机应用技术]