Some Results on Computer Control of Urban Traffic

The problem of controlling vehicle traffic on an urban street network is a particular example of a large scale systems problem, In a reasonably typical network there will be one thousand or more traffic signals, there may be several hundred sensors (vehicle detectors) and the network is typically distributed over several hundred square miles. It is now quite common to control these systems primarily from a central computer. The traffic control problem is modelled by a kind of “store and forward” network in which vehicles are ‘stored’ in queues at the signals and “forwarded’ in platoons. The data from detectors is modelled as a point process that is statistically related to the queues and platoons. It would be very tempting to design optimal feedback control algorithms for these systems. However, the size of the problems makes this impossible. Instead sub-optimal controls are developed. These sub-optimal controls use the minimum error variance estimator of the queues at each intersection as determined from the above model to improve the control provided by several intuitively derived sub-optimal controls. The sub-optimal controls have been tested using a good simulation of urban traffic. The sub-optimal controls give substantially better performance than benchmark open loop controls. Conditions under which the network can be decomposed than into subnetworks without significant degradation in performance are also given.