عنوان مقاله [English]
Estimation (correction) of origin-destination (OD) matrix using traffic counts data is an inexpensive approach to predict travel demand in a transportation network. The general formulation of this problem is a two-level optimization program that solves the matrix estimation problem at the upper level and the traffic assignment problem at the lower level. In congested networks, deterministic user equilibrium (UE) assignment is often used at the lower level. The deterministic approach assumes an identical user perception of network travel times that is not true in reality. This research developed the OD matrix estimation problem (ODMEP) under the stochastic user equilibrium (SUE) condition. In this regard, the SUE assignment with the multinomial logit (MNL) route choice model was applied. MNL is a traditional discrete choice model with a straightforward closed-form choice probability. Also, Spiess’s gradient-based approach, which is efficient in large-scale networks, was used at the upper level. Spiess’s OD estimation model under UE and SUE constraints was implemented on the large-scale Tehran network under different user perception variance represented by the scale parameter in the MNL formula. For comparing the results of the two models (ODMEP with UE/SUE assignment), two scenarios were adopted to generate the initial OD matrix. Furthermore, in each scenario, the scale parameter (θ) with different values was evaluated. Results showed that ODMEP with SUE constraint had better performance in producing link volumes near observed traffic counts than UE-constrained ODMEP. Besides, individual OD demands resulting from the SUE-based model were better fitted with the real OD matrix elements than the UE-based one. However, by increasing the scale parameter θ (decreasing the variance of users’ perception of network travel times), the results of the two methods approach each other. Therefore, if the scale parameter value is known, the SUE-based model would be more accurate and preferable than the UE-based model for low-value scale parameter conditions. In the Tehran network, the SUE-based model can decrease the RMSE of estimated matrix elements more than 10 percent relative to the UE-based model when the scale parameter is less than 0.5.