Modeling and evaluation of integrated dynamic signal and dynamic speed control in signalized networks

A new integrated Dynamic Speed and Dynamic Signal (DSDS) control algorithm for signalized networks is developed in this research. The algorithm is formulated as a dynamic optimization problem with the objective of maximizing the number of vehicles released by the network and minimizing the number of stops in the network. The control algorithm is optimized by Genetic Algorithms (GAs).;The developed DSDS algorithm is applied to signalized networks. The benefits of implementing a DSDS control algorithm on network efficiency are first evaluated through looking at key measures of effectiveness (MOEs). It is demonstrated that the algorithm is able to reduce queues over time, avoid gridlocks, and improve system performance. Vehicle speed profiles under DSDS control and dynamic-signal fixed-speed (DSFS) control are compared to evaluate the advantages of control with dynamic speed on minimizing speed noise and speed variation. DSDS control generates smoother flow profiles by reducing speed noise and speed variation. The comparison provides evidence that implementing DSDS control in signalized networks is an effective way to achieve safer and environmentally friendly signalized network operations. The operational and safety enhancement brought about by the implementation of DSDS varies depending on the levels of driver compliance. The microscopic simulation model VISSIM is used to evaluate the impacts of different levels of driver compliance. Results show that speeding and slow driving each have negative impacts on the performance of DSDS control.;Parallel GAs (PGAs) is investigated and deployed in order to improve computational performance. Both a simple GA (SGA) and island PGAs are used to solve the DSDS control problem, a standard GA-difficult, and a standard GA-easy problem. For all problems, savings in computation resources were realized when PGA was used. The magnitude of improvements brought about by a PGA depended on the difficulty of the problem. An empirical approach is explored to configure Parallel Genetic Algorithms (PGAs) to optimize the DSDS control algorithm developed in this research. Two of the most important island PGA parameters are examined: the number of islands (subpopulations) and the migration rate. The results show 1) increasing the number of subpopulations does not always bring worthwhile savings in time, 2) increasing the number of subpopulations decreases the importance of migration rate, 3) there is an optimal migration rate associated with each number of subpopulations and it is problem-dependent, and 4) PGA configuration and performance with the standard benchmark functions can be used as benchmarks to configure the PGA for problems of unknown complexity, such as the DSDS control algorithm developed in this research. The results suggest that off-line processing may be necessary to ensure optimal performance of the PGA.