This paper investigates the problems which arise when traffic signal settings are "optimised" in response to traffic flows through a network; but the flows themselves are sensitive to these signal settings as drivers seek to minimise their travel times. This interaction has important implications for both how to choose true optimal settings and how to carry out assignment for traffic networks where signal settings are indeterminate. Simulation results obtained using real-life networks with varying levels of traffic congestion are described. The SATURN (Simulation and Assignment of Traffic to Urban Road Networks) network and analysis programs were used for both signal optimisation and assignment. Three signal setting policies were tested: the Equisaturation policy, a Delay minimisation policy (as suggested by Allsop), and the policy Po proposed by Smith (1980) (see IRRD 248771) with the express objective of increasing network capacity while allowing for flow reassignment. The objective of the various computer tests was to find a solution point where junction green times and traffic flows are in equilibrium such that the green times satisfy the control policy, whilst the flows are user optimised. With simple networks, the traditional policies frequently gave rise to multiple equilibria for the combined signal control/assignment process, depending on the initial settings. The Po policy, on the other hand, found unique equilibria, which were also near-optimal at higher congestion levels. Tests on four real-life networks showed a much more diffuse picture. On two smaller networks, Po was more stable than the other two, and performed worse under low congestion, but some 10-12% better than the two more traditional policies when congestion was high. On a third network, all three policies gave virtually the same travel times (but significantly different green time splits?), whilst for the largest Po and Delay minimisation both performed up to 25% better than Equisaturation.
Samenvatting