Short-term Traffic Flow Prediction Based On Tensor Completion

With the development of the society,the proportion of urban population in China is increasing,which brings many challenges to urbanization,and the most important is urban traffic planning.Analyze the current situation of urban traffic in China,among which congestion is the most serious problem,and there are also environmental pollution,parking difficulty and frequent traffic accidents.The reason mainly lies in the increase of private cars,city population growth,network planning unreasonable,public transportation atrophy,too intensive investment and construction,and these problems often lead to a vicious spiral,chain reaction,makes the city traffic congestion and even paralysis.Therefore,if we can accurately predict the traffic volume in a section within a certain period of time,we will be able to prevent and ease the traffic volume at rush hour ahead of time,and fundamentally solve the traffic jams.However,the traffic situation is complex and changeable,and many factors need to be taken into consideration,and there are abundant high dimensional structure information in traffic flow data,It makes it impossible for the traditional prediction model to reach the expected accuracy.For this reason,this paper uses a mathematical model of a high dimensional structure,called a tensor structure.By combining traffic flow data,a high dimensional dynamic tensor model is constructed,and a short-term traffic flow prediction method based on dynamic tensor is realized.The main contents of this paper are the following aspects:First,we introduce the research background and significance of short-term traffic flow prediction,including the research status and development trend of related prediction models at home and abroad,and the research history and application status of tensor models in this paper.Second,the mathematical expression and calculation formula of tensor are introduced in detail,which provides the basis of mathematical theory for the next decomposition model.Then,two commonly used decomposition models of tensor,named CP decomposition model and Tucker decomposition model,are introduced,and the selection and filling theory of factor matrix under two decomposition models are introduced respectively.Third,the real traffic flow data are constructed to the high-order tensor structure,and the correlation and periodicity of traffic flow data in multiple modes are analyzed.Then,the traffic flow data is combined with tensor to fill the theory,and the short-term traffic flow prediction algorithm based on tensor structure is proposed.Fourth,we introduce the detailed process of short-term traffic flow prediction algorithm based on CP and Tucker two decomposition models,including the theoretical basis,the derivation process,the specific implementation steps and the advantages and disadvantages of the algorithm.Finally,the algorithm is analyzed and compared with the actual traffic flow data.It mainly includes the accuracy of the complete data set and the comparison of the algorithm running time,and the algorithm accuracy analysis under the missing data set.In the missing data set,it is also divided into the complete randomness loss and continuity loss of the historical data set.It is found that in these two cases,the proposed Tucker-WGopt has good accuracy.