D. Sun, R.F. Benekohal

Pages: 33-48

Abstract
This paper studies several logit-based route choice models for stochastic user equilibrium problem. Although traditional multinomial logit model has been extensively used for traffic route choice problems, the argument on the violation of Independent Irrelevant Alternatives gives rise to the development of some new logit models including C-Logit Model, Path Size Logit, Generalized Nested Logit and Paired Combinatorial Logit etc. It is apparent that different route choice models provide varying path use probabilities because the similarity factor is treated in different ways. However, the impact of different choice models on equilibrium flow of a network with flow-dependent link cost is not yet clear. The potential overload on some overlapping links can be offset by the monotonically increasing link cost functions to some extent. This paper proposes a MSA-based algorithm which accommodates five route choice models. The equilibrium flow generated by MNL is used as the base flow pattern, the shift on the equilibrium flow caused by extended logit models are investigated with respect to the percentage of change in link flow to study the effectiveness of different models. The saturation levels of each link generated by different models are compared as well. In addition, the impact of extended logit models on the system performance in terms of total travel time is examined and it is found that all extended logit models can lead to some improvement on system performance. PCL produces the most significant change in flow and highest improvement on total travel time. The computational results shows that C-Logit and PSL take similar CPU time to MNL and the CPU time used by PCL is about 3.1 times that of MNL and GNL consumes the longest time to achieve equilibrium.

Keywords: route choice model; stochastic user equilibrium; logit model