Cyclically time-dependent optimization model for traffic signals; interaction between traffic control and demand; exceptional traffic situations
Final Report Abstract
Signal control for urban traffic is an important topic of practical interest. Being able to explicitly model queue built-up at red signals, with the cyclic construction of the time-expanded network, is a break-through of mathematical modeling capabilities. It was shown in this project that the approach can be extended from phase optimization of fixed-time signal plans to the more important green split optimization. An alternative approach are actuated or adaptive signals. As expected they perform worse than fixed time signals in Braess-like situations, where the fixed-time traffic signals can move the routing from Nash equilibrium to system optimum, which is clearly not possible for an actuated or adaptive signal that follows the traffic. We found this not much (if at all) an issue in other practical situations that we investigated. Actuated or adaptive signals also do not depend as much on details of the demand or the intersection geometry. Finally, they provide additional gains when traffic flows fluctuate slightly. Overall, actuated and adaptive signals fared better than initially expected. Robustness was an issue throughout the project. Signal plans might promise much better performance from the mathematical optimization than what they showed in the simulation. For example, the mathematical approach tends to switch signals immediately back to red after the demand has been served, which for any stochastic simulation or real-world situation has the consequence that, because of fluctuations, some vehicle will be stranded at that red light which might have remained green. Without a meaningful extension of these green times, using simulation to evaluate the mathematically found signal plans often displays performances that are far worse than what the mathematical algorithm promised. The project documents several such issues that seem like small differences but lead to large differences in the performance evaluations. By increasing the level of detail in traffic signal modeling (adapting green split, phase order, etc.) the travel time difference between analytical model and simulation increased. This surprising outcome shows that the difference between user equilibria and optimal solutions in the CTEN is not fully understood and calls for further research in the following direction: Which part of the difference between analytical model and simulation is due to the difference between system optimum and user equilibrium, which part results from possibly different equilibria in the dynamic network with traffic signals, and which part is caused by a lack of robustness of the analytical solutions with their sharply timed switching times. Re-routing is rarely considered in the context of signals optimization, but has a strong effect: Gains from optimized signals in urban centers will lead to additional traffic through the urban core, thus worsening environmental impacts. Overall, this implies that signal control schemes need to be evaluated in context, with a large scale model that includes re-routing (and optimally mode and departure time choice), and is able to calculate environmental impacts.
Publications
- Traffic signal optimization using cyclically expanded networks. Networks, 65(3):244– 261, Wiley, 2015
E. Köhler and M. Strehler
(See online at https://doi.org/10.1002/net.21601) - Braess’s paradox in an agent-based transport model. Procedia Computer Science 83, pp. 946–951, 2016
T. Thunig and K. Nagel
(See online at https://doi.org/10.1016/j.procs.2017.05.371) - Optimizing Traffic Signal Timings for Mega Events. In Proceedings of the 16th Workshop for Transportation Modelling, Optimization, and Systems ATMOS 2016, OASICs volume 54, pp. 8:1–8:16, 2016
R. Scheffler and M. Strehler
(See online at https://doi.org/10.4230/OASIcs.ATMOS.2016.8) - Optimizing Traffic Signal Settings for Public Transport Priority. In Proceedings of the 17th Workshop for Transportation Modelling, Optimization, and Systems ATMOS 2017, OASICs volume 59, pp. 9:1–9:15
R. Scheffler and M. Strehler
(See online at https://doi.org/10.4230/OASIcs.ATMOS.2017.9) - The structure of user equilibria: Dynamic coevolutionary simulations vs. cyclically expanded networks. Procedia Computer Science 109, pp. 648–655, 2017
T. Thunig and K. Nagel
(See online at https://doi.org/10.1016/j.procs.2017.05.371) - Equilibria in routing games with edge priorities. In Proceedings of the 14th Conference of Web and Internet Economics WINE 2018, LNCS 11316, pp. 408–422, 2018
R. Scheffler, M. Strehler, and L. Vargas Koch
(See online at https://doi.org/10.1007/978-3-030-04612-5_27) - Adaptive traffic signal control for real-world scenarios in agent-based transport simulations. Transportation Research Procedia 37, pp. 481–488, 2019
T. Thunig, N. Kühnel and K. Nagel
(See online at https://doi.org/10.1016/j.trpro.2018.12.215) - Effects of user adaption on traffic-responsive signal control in agent-based transport simulations. 6th International Conference on Models and Technologies for Intelligent Transportation Systems MT-ITS’19, 2019
T. Thunig and K. Nagel
(See online at https://doi.org/10.1109/MTITS.2019.8883373) - Optimization and simulation of fixed-time traffic signal control in real-world applications. 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models, Methodologies and Applications ABMTRANS’19, Procedia Computer Science, 151(2019)
T. Thunig, R. Scheffler, M. Strehler, and K. Nagel
(See online at https://doi.org/10.1016/j.procs.2019.04.113) - Traffic Signal Optimization: combining static and dynamic models. Transportation Science 53(1), pp. 21–41, INFORMS, 2019
E. Köhler and M. Strehler
(See online at https://doi.org/10.1287/trsc.2017.0760)
