The network design problem: The traffic-responsive signal control scheme

This thesis concerns Bilevel Nonlinear Programming (BNP) and its application to the Network Design Problem (NDP): the Traffic-Responsive Signal Control (TRSC) scheme. The TRSC scheme considers the effect of signal controls on traffic flows in the standard traffic assignment problem through the integration of traffic control and traffic assignment submodels. Alternative problem formulations and available solution algorithms are briefly reviewed. Eight models based on two traffic signal timing policies, two different treatments of two-way stop intersections, and two different weights for turning movements are thus formulated. Among them four models are illustrated and compared. The iterative optimization and assignment (IOA) solution algorithm is suggested for solving a large-scale real network: the Chicago Area Transportation Study network. This real network contains 413 zones, 8408 nodes and 26719 links, of which a subregion consisting of 271 zones, 1059 nodes, and 2898 links is of primary interest. All the discussions are restricted to the subregion unless otherwise stated. The entire study area may be the largest network that has been tested for any type of the TRSC scheme until now.It is observed that the solutions for the four models are stabilized after six iterations of the TRSC scheme. It is also found that the difference between the delay minimization and the equal degree of saturation traffic signal timing policies tested is not significant. For simplicity, the equal degree of saturation traffic signal timing policy may be safely used to approximate the delay minimization traffic signal timing policy. The treatment of two-way stop intersections by a simplified 1985 HCM method does not make much difference in the objective values as well. The models' performance and benefit that could be obtained from the TRSC scheme were extensively discussed by eight network performance measures and comparative studies.The thesis also describes our experience with code optimization on the Alliant FX/8, a mini-supercomputer. The parallel nature of the problem, mainly in the minimum path search and all-or-nothing assignment algorithms, is explained. Moreover, the computer memory requirement is not necessarily increased for the optimized code. The importance of adopting the parallel computing technique was justified by the case study and may be emphasized more by a new research area--vehicle navigation guidance system.U of I OnlyETDs are only available to UIUC Users without author permissio