Research On Traffic Data Recovery Based On Tensor Completion And Coupled Matrix And Tensor Factorizations

Smart transportation is an important part of smart city,it mainly studies how to combine big data,Internet of Things,cloud computing and artificial intelligence,to build up the more prompt,efficient and powerful transportation management system.Traffic data is the basis of transportation management system.Unfortunately,missing traffic data are inevitable due to equipment failures and extreme weather,which seriously restricts the progress of smart transportation construction.Therefore,in recent years,missing traffic data recovery has become a research hotspot in the field of intelligent transportation.Based on tensor completion and CMTF(coupled matrix and tensor factorizations),this dissertation mainly studies two problems existing in the field of missing traffic data recovery,including traffic data modeling framework and missing traffic data recovery model in extreme conditions.The main works of this dissertation can be summarized as follows:Firstly,this dissertation introduces the background,significance and research progress of missing traffic data recovery,then the basic concepts,classic theories and methods of tensor theory and missing traffic data recovery are presented.Secondly,in order to solve the problem that how to find and use the most relevant feature to model the traffic data,this dissertation proposed a traffic data modeling framework based on traffic flow data,and systematic study the influence of different data modeling methods on the performance of tensor-based traffic data recovery method.The experimental results showed that utilize the multi-dimensional internal correlation of traff-ic data reasonably can significantly improve the accuracy of missing data recovery.Finally,for extreme conditions,a novel traffic data recovery model,which is based on coupled matrix and tensor factorizations is proposed.Experiments demonstrate that the proposed model achieves a better performance than the state-of-the-art traffic data recovery methods,even in extreme conditions,the proposed model still keep the good stability and effectiveness.