Investigation of Stochastic Network Loading Procedures

Abstract

The network loading process of stochastic traffic assignment is investigated. A central issue in the assignment problem is the behavioral assumption governing route choice, which concerns the definition of available routes and the choice model. These two problems are addressed and reviewed. Although the multinomial logit model can be implemented efficiently in stochastic network loading algorithms, the model suffers from theoretical drawbacks, some of them arising from the independence of irrelevant alternatives property. As a result, the stochastic loading on routes that share common links is overloaded at the overlapping parts of the routes. Other logit-family models recently have been proposed to overcome some of the theoretical problems while maintaining the convenient analytical structure. Three such models are investigated: the C-logit model, which was specifically defined for route choice; and two general discrete-choice models, the cross-nested logit model and the paired combinatorial logit model. The two latter models are adapted to route choice, and simple network examples are presented to illustrate the performance of the models with respect to the overlapping problem. The results indicate that all three models perform better than does the multinomial logit model. The cross-nested logit model has an advantage over the two other generalized models because it enables performing stochastic loading without route enumeration. The integration of this model with the stochastic equilibrium problem is discussed, and a specific algorithm using the cross-nest logit model is presented for the stochastic loading phase.