Estimation of trip matrices from traffic counts : an equilibrium approach

• Main Author: Oh, Jae Hak
• Format: Dissertation
• Language: English
• Published: University of London 1992

In urban traffic management and planning, an important problem is how to obtain estimates of origin-destination (O-D) trip matrices using low-cost data such as traffic counts. Although conventional methods using the data from direct surveys can be used to estimate trip matrices, they appear to be inaccurate and expensive. By contrast, the use of traffic counts is attractive, as it is less expensive and more practical. The main objective of the research reported in this thesis is to develop new methods for estimating trip matrices from traffic counts when congestion effects in networks are considered. The problem and existing methods including the sequential solution method used in the ME2 model are reviewed. A new formulation is given for the problem which solves the two sub-problems of entropy maximization and equilibrium traffic assignment simultaneously. It allows modelled link flows to be constrained so as to equal observed ones without link assignment proportions of the trips. A simultaneous solution method is presented for this new formulation. To reduce the considerable computational burden in solving the problem, a heuristic method has been developed which uses a linear approximation fitted by regression to the equilibrium link flows. Extrapolation and perturbation methods have also been used to speed up the solution process. However, the simultaneous solution method appears to be impractical for use in large networks because of the heavy computational demand. As an alternative, an improved sequential solution method is proposed which uses a penalty function method. This method approximates a solution by solving a sequence of problems, while fixed link assignment proportions are used. The performance of the proposed methods has been tested and compared with that of the existing sequential ME2 method using both small example networks and a larger real network. The results show that the simultaneous method works well and that it performs better than the existing sequential method or the improved sequential method. The improved sequential method is also shown to perform closely to the simultaneous one. Some practical implications of the new methods including the robustness of the solutions and the increased computational burden are discussed and they are also compared with those of the sequential solution method. The conclusions from the main findings of the research are drawn and a number of suggestions for further study are given.