Study On Traffic Assignment Models And Algorithms Based On Extended Logit Models

With the growing contradiction between the supply and demand of urban traffic in China, the traffic problems have become one of the major bottlenecks which restrict the urban development. In order to solve the traffic problems and improve the traffic situation, the government departments consider the traffic supply and demand simultaneously, and increase the construction intensity of the traffic infrastructure and improve the level of travel demand management continuously. To ensure the traffic planning and management measures to achieve the best effects, it needs to adopt the advanced traffic planning theories and methods to assist the decision making processes. The research of traffic assignment models is not only an important part of modern traffic planning theory, but also a core technology in the studies of urban road network design problem (URNDP) and congestion road pricing (CRP). In this context, this dissertation mainly focuses on urban traffic networks, and studies on the traffic assignment models which could describe the travelers'route choice behaviors more accurately and applies them in the studies of network design problem and congestion road pricing. And this make the studies of this dissertation form a complete system to provide supports for the decision making of traffic managers and planners. The main studying aspects of this dissertation are as following:(l)In order to overcome the independence of irrelevant alternatives (?A) property which existed in the traffic assignment models based on the traditional Logit model, the route perception Logit model (RPL-O) which only considered the route overlapping problem and the path size correction Logit model (PSCL) are used in this dissertation, and the stochastic user equilibrium models based on RPL-O and PSCL models were constructed respectively, and the equivalence of the proposed models and the uniqueness of the solutions were proved; the path-based method of successive averages (MSA), method of successive weighted averages (MSWA), and self-regulated averaging method (SRA) algorithm were designed to solve the two proposed stochastic user equilibrium models; through the numerical experiment, the traffic assignment results between the new models and the traditional Logit-based stochastic user equilibrium model were compared in this dissertation and the efficiencies of the three proposed algorithms were analyzed, and the results showed that SRA algorithm was the most efficient algorithm among them. (2)The forms of selection probability and the methods to determine the parameters of C-Logit, path size Logit, generalized nested Logit, paired combinatorial Logit were described; the multi-class multi-criteria stochastic user equilibrium models based on the C-Logit, path size Logit, RPL-O, path size correction Logit, generalized nested Logit, paired combinatorial Logit models were constructed, and the equivalence of the six proposed models and the uniqueness of solutions were proved; the representation of road network structure and calculation methods of the route selection probability of different models were compared and analyzed; the common path-based SRA algorithm was designed to solve the proposed models; through the numerical experiment, the traffic assignment results between the new models and Logit-based multi-class multi-criteria stochastic user equilibrium model were compared, and the impacts of model parameters on the assigned flows were analyzed.(3)The deficiencies of the existing studies on continuous network design were analyzed, and the minimum summation of the total travel costs and investment amount was used in the upper level objective function, and the two models proposed in chapter three and the other four extended Logit-based stochastic user equilibrium models were adopted in the lower level models, then the six new continuous traffic network design models were constructed; the combination algorithm which combined the genetic algorithm and SRA algorithm was designed; through the numerical experiment, the effectiveness of the algorithm was demonstrated and the differences of the network design schemes between the extended Logit-based stochastic user equilibrium models and the traditional deterministic user equilibrium model used as the lower level model were analyzed, the changes in the network design results were analyzed when the model parameters varied.(4)By analyzing the deficiencies of the traditional second-best congestion pricing studies, the significance to consider the multi-class multi-criteria and the use of extended Logit-based traffic assignment models on the studying of congestion pricing was indicated; the minimum total travel time was used as the upper level objective function, and the extended Logit-based multi-class multi-criteria stochastic user equilibrium models were adopted in the lower level models, six new second-best congestion pricing models were built; the combination algorithm which combined the simulated annealing algorithm and SRA algorithm was designed; through the numerical experiment, the effectiveness of the combination algorithm was tested and the differences of the congestion pricing schemes between the extended Logit-based multi-class multi-criteria stochastic user equilibrium models and the multi-class multi-criteria deterministic user equilibrium model used as the lower level model were analyzed, the changes in the congestion pricing results were analyzed when the model parameters varied.