Adaptive traffic control system (ATCS) aims at controlling the imminent traffic, which is yet to arrive and hence not known perfectly. An ATCS can use either historical data based on time of day or day of week, or real-time detected data, to formulate control strategies, in the hope that the current or historical arrival profile will remain representative for the upcoming situation....[ Read more ]

Adaptive traffic control system (ATCS) aims at controlling the imminent traffic, which is yet to arrive and hence not known perfectly. An ATCS can use either historical data based on time of day or day of week, or real-time detected data, to formulate control strategies, in the hope that the current or historical arrival profile will remain representative for the upcoming situation.

Nevertheless, a reliable and efficient ATCS should posses a traffic flow / control model, which can model traffic dynamics and optimize traffic signal timings accurately under all traffic conditions, which include both under-saturated and over-saturated situations. For this purpose, as a first step Lo (1999, 2001) developed a cell-based dynamic signal control formulation designated as Dynamic Intersection Signal Control Optimization (DISCO). Previous studies demonstrate that DISCO shows promise as a new approach for demand-responsive traffic control as it outperforms TRANSYT by more than 30% in terms of overall delay reduction when applied to Hong Kong network scenarios (Lo et al., 2001).

In this thesis we mainly focus on exploring ways to improve the computational speed of DISCO and investigating the effects of traffic flow data quality on performance of DISCO.

Lo (1999, 2001) used mixed-integer program (MIP) and genetic algorithm (GA) to solve for the optimal timing plan for DISCO. For large networks, MIP requires extensive computational time and effort to solve for an optimal solution. Compared with MIP, GA is able to solve the problem more efficiently. For real-sized problems, however it still requires a long computation time to solve for DISCO even with GA. In order to be applied for real-time traffic control, DISCO would have to generate its optimal or quasi-optimal timing plans more quickly. To rectify this shortcoming, we introduce a new proposed signal optimization algorithm by combining GA with Frank-Wolfe Combination algorithm to search for the optimal timing plan.

To increase the control flexibility of DISCO for dynamic traffic, we also propose three distinctive control strategies (Lo and Chow, 2002), ranging from the conventional "fixed green splits in fixed cycles" (FGFC) plans, to "time-variant or variable green splits in fixed cycles" (VGFC) strategy, and eventually "time-variant or variable green phases, variable cycle" (VGVC) strategy. In terms of control flexibility, VGVC includes VGFC as a special case, which in turn includes FGFC as a special case.

In addition, performance of ATCS would depend heavily on the accuracy of prediction of the traffic flowing into the network. In the second part of the thesis, we investigate the relationship between various control strategies, traffic data Resolution, and prediction accuracy (notated in short as the CRA relationship) through an extensive simulation of scenarios in Hong Kong with DISCO. The major findings include: (i) the importance of resolution outweighs that of error; (ii) dynamic timing plans generally outperform time-invariant timing plans; (iii) up to a certain extent, overestimated predictions lead to better results than underestimated predictions.