基于MDS-LSTM的航路网络短时交通流预测  

Short Term Traffic Flow Prediction of Air Route Network Based on MDS and LSTM

作　　者：李桂毅[1] 吕晓扬 李沛谦 张洪海[1] LI Gui-yi;LYU Xiao-yang;LI Pei-qian;ZHANG Hong-hai(Nanjing University of Aeronautics and Astronautics,Nanjing 211000,China)

机构地区：[1]南京航空航天大学,江苏南京211000

出　　处：《航空计算技术》2022年第1期60-64,共5页Aeronautical Computing Technique

基　　金：国家重点研发计划项目资助(2018YFE0208700)。

摘　　要：针对航路网络交通流时间序列预测问题,提出基于多维标度法与长短时深度神经网络的航路网络短时交通流预测方法,提升航路网络交通流预测的精度。依据航路网络航迹数据,提取路网航段交通流时间序列数据,并进行降噪滤波处理;依据路网航段交通流相关性,利用多维标度法划分预测航段组合;构建基于长短时深度神经网络的航路网络航段短时交通流回归预测模型,并进行神经网络的调参与训练,实现航路网络航段短时交通流实时预测。实验结果表明:通过引入长短时多层深度神经网络构建的预测模型能更好地拟合航路网络交通流演变规律,预测平均绝对误差均小于0.1,优于随机森林等机器学习模型,预测精度及稳定性较好。Aiming at the time series of air route network traffic flow,a short term traffic flow prediction method of air route network based on multi dimensional scaling method and LSTM is proposed to improve the accuracy of air route network traffic flow prediction.Firstly,the air route network segment traffic flow time series data are extracted and denoised according to the air route network flight trajectory data.Secondly,according to the traffic flow correlation of air route network segments,the predicted air segment combination is divided by multi dimensional scaling method.Finally,a short term traffic flow regression prediction model is constructed based on LSTM of air route segment.The parameter adjustment and learning training of LSTM is carried out to realize short term traffic flow prediction of air route network.The experimental results show that the prediction model can better fit the evolution law of route network traffic flow by introducing LSTM multi layer depth neural network.The average absolute error of prediction is less than 0.1,which is better than machine learning models such as random forest,and the prediction accuracy and stability are good.

关 键 词：航空运输 交通流预测 航路网络 长短时深度神经网络 多维标度法 

分 类 号：V355[航空宇航科学与技术—人机与环境工程]