Study On Road Network Traffic Flow Data Imputation Method Based On Low-rank Matrix Completion Model

Intelligent transportation system(ITS)is one of the effective ways to alleviate traffic congestion and improve the efficiency of road traffic.However,due to the delay of network transmission and the failure of the detectors,there are many missing values in the traffic flow data collected in ITS,which makes it very difficult to implement some applications,such as traffic flow prediction and vehicle path planning.Currently,imputation of traffic flow missing values has become an important research topic in the field of intelligent transportation,which has attracted much attention from both domestic and foreign scholars.Recently,the low-rank matrix completion(LRMC)model based traffic flow missing value imputation has become a hot research topic.However,the current research directly applies LRMC model to the collected traffic flow data with missing values,which does not fully take into account the inherent characteristics of traffic flow data,thus reducing the performance of missing value imputation.Based on low-rank matrix completion model,the imputation problem of road network traffic flow data is systematically studied in this paper.The main works include:(1)It analyzes the traffic flow characteristics of road network and the causes of the missing data of traffic flow.It also introduces three typical missing patterns,which are MCAR(missing completely at random),MAR(missing at random),and MIXED(MCAR and MAR each accounts for 50%)?(2)The missing data imputation problem of road network traffic flow based on low-rank matrix complement is described,and then a common optimization algorithm,termed as singular value threshold iteration(SVT),is introduced in detail.(3)We consider that the heterogeneity of traffic flow data spatial-temporal correlation should be taken into account when applying low-rank matrix completion for traffic flow data.We propose a missing value imputation algorithm called CLRMC-EN which is based on the correlation of traffic samples and ensemble learning.The detailed flowchart and the time complexity of the algorithm are described.The performance of the algorithm is evaluated on a publicly available traffic flow data set collected at Portland,Oregon,USA.The simulation results show that CLRMCEN is significantly better than LRMC and other common imputation algorithms.(4)In view of the high time complexity of CLRMC-EN algorithm for large-scale network,a missing value imputation algorithm based on clustering algorithm and least square ensemble learning(HCLRMC-EN)is proposed.The principle and procedure of the algorithm are explained in detail.The simulation results on the traffic flow data collected from the same road network but with more sensors show that the imputation performance of HCLRMC-EN algorithm reaches 96% of CLRMC-EN algorithm,but runs 19 times faster,which is more suitable for large-scale road network.