| Under the background of the continuous growth of domestic vehicle ownership and the number of drivers,the urban transportation system is under enormous pressure.The existing traffic resources cannot meet the travel demand of residents,the supply and demand of traffic resources are seriously unbalanced,and the problem of traffic congestion has become prominent.How to improve the operational efficiency of the transportation system has become the primary goal of urban planning.The travel characteristics of urban residents are key factor in evaluating the rationality of urban spatial structure and the connectivity of urban transportation networks.Mastering the temporal and spatial characteristics of residents’ travel and mining residents’ travel demand are of practical significance for providing theoretical support for urban traffic planning,management and smart city construction.As an important part of the Intelligent Transportation System,the license plate recognition technology has been widely used in many cities.However,due to transmission line failures,communication failures and other reasons,the system cannot detect and recognize the vehicle at a certain time or period,resulting in some data missing.The problem of data missing is common and unavoidable in data-driven intelligent transportation systems.Complete and accurate traffic flow data is the basis for mining residents’ travel characteristics.Meanwhile,the travel characteristics of urban residents are affected by factors such as the operation status of intersection traffic flow,land type and residents’ travel purpose.The multi-factor influence degree analysis can identify important factors that affect residents’ travel characteristics,and the results reflect the value distribution law of residents’ travel characteristics under different combinations of factors,which has high application value.Therefore,the core research content of this thesis includes the following three parts: imputing intersections traffic flow data,mining the temporal and spatial characteristics of vehicle travel,and analyzing the multi-factor influence degree of vehicle travel characteristics.Firstly,this thesis proposes a rank-adaptive Bayesian tensor decomposition model to impute missing traffic flow data.The model builds data structures based on a tensor model,uses Bayesian model to place flexible priors and hyper-priors on the parameters and hyper-parameters of the tensor decomposition model,and solves the rank selection problem of the tensor decomposition model through a rank-adaptive algorithm.Subsequently,the thesis examines the accuracy of the model’s data imputation under different tensor data structures,different data losing methods,different losing rates,and different ranks of tensor decomposition.The experimental results verify the effectiveness and accuracy of the model in imputing traffic flow data.Secondly,based on the non-negative tensor CP decomposition model,this thesis reveals the distribution law of the spatio-temporal characteristics of vehicle travel in Changsha from the perspective of data mining.Utilizing factor matrix,three basic travel modes are extracted from collective movement in the time dimension,and the spatial distribution and hotspot areas of the vehicle’s basic travel modes are identified in the spatial dimension.Finally,this thesis builds a method framework combining clustering algorithms,principal component analysis and tensor decomposition and reconstruction model,which is used to study the influence degree analysis of factors that affect vehicle’s travel.Traditional factor influence analysis methods suffer from the problems of excessive number of factors,collinearity interference among factors,and difficulty in applying to large-scale data sets.This is one of the few studies where the tensor model is used in multi-factors influence degree analysis,and it’s also one of the few studies used in multi-factors influence degree analysis in the transportation field.There are 23 figures,16 tables and 79 references in this thesis. |
PDF Full Text Request