This paper examines the suitability of implementing a dynamic signal control system in Hong Kong based on the cell-transmission model (CTM). This model was developed by Daganzo (1994) at UC Berkeley and is a convergent numerical approximation to the Lighthill and Whitham (1955) and Richards (1956) model (also known as the kinematic wave model) of traffic flow. As such, the CTM model covers the full range of fundamental density-flow-speed relationships, automatically capturing the macroscopic features of traffic, including shockwaves, queue formation and queue dissipation, in both congested and uncongested regimes....[ Read more ]

This paper examines the suitability of implementing a dynamic signal control system in Hong Kong based on the cell-transmission model (CTM). This model was developed by Daganzo (1994) at UC Berkeley and is a convergent numerical approximation to the Lighthill and Whitham (1955) and Richards (1956) model (also known as the kinematic wave model) of traffic flow. As such, the CTM model covers the full range of fundamental density-flow-speed relationships, automatically capturing the macroscopic features of traffic, including shockwaves, queue formation and queue dissipation, in both congested and uncongested regimes.

We developed a CTM-based traffic simulator that captures the features of Hong Kong streets, including turning movements, special turning lanes and signalized merges. Using field data gathered from a small urban street network in Hong Kong, we find that the CTM delay estimates are at least as good as TRANSYT estimates for non-boundary links under uncongested conditions. Under congested conditions, the CTM estimates are noticeably better than TFLANSYT's estimates.

However, accurate simulation of traffic conditions alone is not enough to implement a demand-responsive signal timing system; one must be able to use these predictions to quickly determine the plan that best suits current conditions. We proposed two genetic-algorithm-based methods (Net-GA and S-GA) for determining quasi-optimal, variable-cycle signal timing plans and compare their performance against TRANSYT and against each other. Both the Net-GA and S-GA methods arrive at signal timing plans that perform better than those produced by TRANSYT. In uncongested conditions, the S-GA method arrives at a better solution in less CPU time than the Net-GA method, while the opposite is true for congested conditions. The resulting quasi-optimal signal-timing plans utilize variable cycle lengths and deploy a dynamic strategy to minimize delay model that clears out "greenbands" for paths through the network. The order and width of these green bands changes automatically in response to existing traffic conditions. This paper concludes with suggestion for further research.