Transport Network Flow Evolution Models With Bayesian Inference

Neither the optimization of urban road network dynamic control measures,nor the design of intelligent transportation systems,can be independent from the simulation about traffic flow dynamic evolution process.Transport network flow evolution model can be regarded as the generalization of traditional traffic equilibrium assignment method,it considers the traveler's individual experience learning and route-switching behavior,and therefore can describe the corresponding network traffic day-to-day dynamic evolution at an aggregate level.Due to the advantages of dynamic evolution model,it has become a hot area of research in the last three decades.Nevertheless,most of the existing models are not unified in the aspect of research perspective and modeling method,and their relations are still not clear.Moreover,a lot of research work only focus on model theoretical analysis,and usually ignore model parameter estimation and their actual application problem.This dissertation aimed at establishing both deterministic process(DP)and stochastic process(SP)dynamic evolution model under a unified framework,analyzing and comparing their respective characteristics and applicable conditions.Meanwhile,this dissertation designed statistical inference method for parameter estimation of the proposed dynamic evolution models.This dissertation studied several theoretical issues related to the transport network flow evolution models,and made efforts to develop the complete procedure as“Modeling Establishment-Characteristic Analysis—Parameter Estimation".The main research results are summarized as below:(1)The construction of research system and the establishment of research method.Two types of dynamic evolution models,including the DP one and the SP one,were selected as the research objects in this dissertation.A generalized model framework was put forward by studying the traveler's individual experience learning and route-switching behavior.The nonlinear dynamics and stochastic process(Markov chain)were adopted as the methodology foundations of these two types of evolution models,according to their respective theoretical characteristics.Furthermore,to comply with the accessible and reliable principles,the day-to-day observations of link traffic volumes were regarded as the data sources for model parameter estimation.And in view of the likehood information contained in the observation data,the Bayesian inference was determined as the parameter estimation method for the proposed evolution model.(2)The establishment of DP evolution model and the investigation of its theoretical characteristics.According to the proposed model framework,a DP dynamic evolution process was established to describe the day-to-day“mean”change trend of the road network traffic flow.Meanwhile,the nonlinear dynamical system theory was applied to investigate several theoretical characteristics of the proposed DP model,including fixed point and its existence,uniqueness as well as stability.(3)The extension of DP evolution model on stochastic transport network.Under the situation of stochastic road capacity degradations,an extended DP model was established,to capture the day-to-day evolution process of travelers' risk-taking route choices.And the traffic prediction information,which can be provided by ATIS,was also introduced into the dynamic system to investigate its influence on the whole transport network system.In the extended model,the concept of 'Travel Time Budget'(TTB)was adopted to reflect the route disutility evaluated by travelers,and the consistent traffic information prediction mechanism was also designed to reflect ATIS traffic information release strategy.Both the theoretical analysis and the numerical experiments of the extended DP model was conducted,to verify the influences of travelers' risk attitudes and ATIS traffic predictive information on the day-to-day network flow evolution process.(4)The establishment of the SP evolution model and the construction of its implementation method.The inter-day OD traffic demand fluctuations and individual behavior subjective randomness were further analyzed in the dynamic model framework,and then a SP dynamic evolution model was proposed to describe the influences of these two random factors on the day-to-day road network traffic flow evolution process.Besides,the regularity condition for the proposed SP model was introduced,and an implementation method based on Monte Carlo simulation technology was designed.(5)The relationship analysis of the DP and SP evolution models.The asymptotic result of road network traffic flow stationary distribution was presented,and the relations among the stable results of DP and SP evolution models as well as the stochastic user equilibrium(SUE)state were investigated.In addition,an asymptotic analysis of the SP model was conducted to check its relationship with the DP countpart.According to the theoretical analysis results of these two kinds of models,the application conditions of these two models were also discussed.(6)The establishment of model parameter estimation method and the design of sampling algorithm.A kind of Baysian inference model was constructed,to estimate the various parameters of the transport network flow evolution models,by inputing the day-to-day observations about link traffic counts.On this basis,a specific Markov Chain Monte Carlo(MCMC)sampling algorithm was designed,for calculating the posteriori distribution of parameters in the Bayesian model.