01557013

Component

In demand estimation problems the number of possible solutions generally increases with the size and the complexity of the network. Hence, it is relevant introducing procedures to reduce the solution space without increasing the problem complexity. Driven by this motivation, this work proposes a procedure to simplify the demand estimation problem in the dynamic case while guaranteeing reliable solutions and without increasing problem complexity. The procedure does not claim to be an alternative to existing theoretical estimation approaches, but focuses on extending and testing practical solution algorithms based on previous models developed by the authors, in order to improve their applicability, especially in cases where a reliability of an a-priori estimate of the demand cannot be guaranteed. In fact, the assumption often made by researchers about its reliability is not always true. Thus, for dealing with this occurrence a Two-Steps approach, previously proposed by the authors, is extended in this paper: starting from a wrong “a priori” matrix, in the first step it aims at estimating the proper level of demand generated by any traffic zone in each time period while accurate demand distributions between the different origin-destination (OD) are estimated in the second one. Tests carried out show that a more reliable estimation of the demand over time and space is achieved. Main contributions of the new approach are a) a considerable reduction of the variance in calibration parameters, implying that more robust and accurate solutions have been obtained; b) the demand estimation procedure is not network size dependent. Further a new variant of the widely adopted Simultaneous Perturbation Stochastic Approximation (SPSA) algorithm is proposed in the second step, where an opportune subset of the OD flows is perturbed at once. Performed experiments show that the proposed procedure is able to obtain results as accurate as those of the conventional SPSA reducing the number of variables to be used in each iteration up to 50%.

This paper was sponsored by TRB committee ADB30 Transportation Network Modeling.

01550057

15-4009

English

17p

2015

Transportation Research Board 94th Annual Meeting

Digital/other

Figures; References; Tables

Algorithms; Calibration; Dynamic models; Estimating; Origin and destination; Reliability; Travel demand

Planning and Forecasting; Transportation (General); I72: Traffic and Transport Planning

Transportation Research Board Annual Meeting 2015 Paper #15-4009

TRIS, TRB, ATRI

Dec 30 2014 1:18PM