摘要
对实时车辆流量、平均车道占用率等各种交通监控数据的完整获取,是建设智能交通系统、提高交通管理运行效率的重要基础。文章提出一种融合类信息的函数型矩阵填充方法(Functional Matrix Completion Method with Class Information,CFMC)。在函数型数据分析框架下,基于非负矩阵分解构造函数型矩阵填充模型,在此基础上通过聚类划分引入样本类信息,借助类内样本相关性插补缺失值,并采用自加权集成学习算法动态赋权计算得到最终插补值。在公共交通数据集PeMS上进行插补实验,结果表明:当缺失率为15%~70%时,CFMC方法相较于K近邻算法、MICE、PACE等10种插补方法,均方根误差(RMSE)、平均绝对误差(MAE)和平均绝对百分比误差(MAPE)分别降低了10.75%~81.69%、0.34%~84.48%和12.5%~81.08%,且耗时可控。所提CFMC方法插补精度高、鲁棒性好,能够保证插补的有效性和准确性。
The complete acquisition of real-time vehicle flow,average lane occupancy and other traffic monitoring data is an important basis for the construction of intelligent transportation systems and the improvement of traffic management efficiency.This paper proposes a Functional Matrix Completion Method with Class Information(CFMC).In the framework of functional data analysis,a functional matrix completion model is constructed based on nonnegative matrix factorization.On this basis,the sample class information is introduced by clustering division;the missing values is imputed by intra-class sample correlation,and the final imputation values is calculated by dynamic weight reweighting based on self-weighted ensemble learning algorithm.The imputation experiment is carried out on the public transport data set PeMS,and the results show that when the missing rate is 15%~70%,compared with K-nearest neighbor algorithm,MICE,PACE and other 10 imputation methods,the root mean square error(RMSE),mean absolute error(MAE)and mean absolute percentage error(MAPE)of CFMC method are reduced by 10.75%~81.69%,0.34%~84.48%and 12.5%~81.08%,respectively,with the time consumption controllable.The proposed CFMC method has high imputation precision,greatly robustness,able to guarantee the effectiveness and accuracy of imputation.
作者
高海燕
马文娟
薛娇
Gao Haiyan;Ma Wenjuan;Xue Jiao(School of Statistics and Data Science,Lanzhou University of Finance and Economics;Gansu Key Laboratory of Digital Economy and Social Computing Science,Lanzhou 730020,China)
出处
《统计与决策》
CSSCI
北大核心
2023年第23期40-45,共6页
Statistics & Decision
基金
国家社会科学基金资助项目(19XTJ002)
甘肃省自然科学基金资助项目(23JRRA1186)
甘肃省优秀研究生“创新之星”项目(2023CXZX-703)
兰州财经大学科研项目(Lzufe2023C-005)。
关键词
函数型数据分析
非负矩阵分解
矩阵填充
交通流量
缺失插补
functional data analysis
nonnegative matrix factorization
matrix completion
traffic flow
missing imputation
