顾及不同缺失率的LSTM城市交通速度插值方法  

A long short-term memory neural network interpolation method for urban traffic speed considering different missing rates

作　　者：罗升 龚循强[1,2] 夏元平 汪宏宇 张瑞[3] LUO Sheng;GONG Xunqiang;XIA Yuanping;WANG Hongyu;ZHANG Rui(Key Laboratory of Mine Environmental Monitoring and Improving around Poyang Lake of Ministry of Natural Resources,East China University of Technology,Nanchang 330013,China;School of Surveying and Geoinformation Engineering,East China University of Technology,Nanchang 330013,China;State-Province Joint Engineering Laboratory of Spatial Information Technology for High-speed Railway Safety,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]东华理工大学自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室,南昌330013 [2]东华理工大学测绘与空间信息工程学院,南昌330013 [3]西南交通大学高速铁路安全运营空间信息技术国家地方联合工程实验室,成都611756

出　　处：《测绘工程》2025年第2期16-23,共8页Engineering of Surveying and Mapping

基　　金：国家自然科学基金资助项目(42101457,42174055);自然资源部环鄱阳湖区域矿山环境监测与治理重点实验室开放基金重点资助项目(MEMI-2023-01)。

摘　　要：针对交通速度值存在随机缺失,不同缺失率数据特征不同的情况,文中将真实交通速度值划分为训练集和测试集进行实验,在测试集数据缺失10%~100%的条件下,基于长短期记忆神经网络(LSTM)的插值方法与基于统计学插值方法中的均值法、最后观测值法,以及基于浅层机器学习插值方法中的支持向量回归法(SVR)、随机森林法(RF)进行对比分析。实验结果表明,随着缺失比例的增加,不同方法的插值结果相差逐渐增大,相比于次优插值方法,基于LSTM的插值方法在各种缺失率下的平均绝对误差、均方根误差和平均绝对百分比误差分别平均提升18.232%、15.714%和17.195%。从而证明基于LSTM的插值方法精度更高,能够有效增强交通速度序列的完整性。True traffic speed data is divided into training and testing sets for experiments and used to compare and analyze interpolation methods based on long short-term memory neural networks with statistical interpolation methods such as Mean and Last Observation Carried Forward,as well as machine learning methods such as Support Vector Regression and Random Forest under the condition of data missing 10%~100%in the test set,which addresses the situation of traffic speed data has random missing data and different missing rates have different data characteristics.The experimental results show that as the proportion of missing values increases,the difference between the results of different methods gradually increases.Compared with the suboptimal interpolation method,the mean absolute error,root mean square error and mean absolute percentage error of the LSTM-based interpolation method are increased by 18.232%,15.714%and 17.195%,respectively.The interpolation method based on long short-term memory neural network has higher interpolation accuracy,and the interpolation results are closer to the true values,which can effectively improve the quality of traffic flow data.

关 键 词：智能交通 随机缺失 长短期记忆神经网络 交通速度插值 

分 类 号：P225[天文地球—大地测量学与测量工程]