Most isolated signal controlled road junctions in Britain now operate under vehicle-actuated control. With the advent of microprocessor controllers, interest is turning to increasingly sophisticated forms of vehicle-actuated control. In this paper, a novel approach to the optimisation of traffic signal control in real time is described, in which the current state of traffic at the junction, including queue lengths, is estimated using a model of traffic behaviour. Conventional inductive loop detector information is used in an optimal filtering framework to update and correct the estimates of system state yielded by the traffic model, as well as the model parameters themselves. The traffic model is then used by the optimiser within a dynamic programming framework to evaluate the likely consequences of alternative courses of action. In this framework, the observations are used to counteract the tendency for the predictions of the traffic model to drift away from the true state of the system. This is possible although the observations themselves may relate only indirectly to the state of the system, provided that the condition of observability is fulfilled. Development of the filtering and optimisation modules is taking place on a Sun workstation. These modules are interfaced with a simulation of traffic at the junction which is sufficiently detailed to provide realistic detector output and measures of performance. To achieve this, a vehicle-following sub-model has been adopted which is solved in discrete time. An event-based framework is used to simulate the control and operation of the junction. The computational demands of such a detailed simulation prohibit its use for optimisation: here its purpose is to facilitate development of the filtering and optimisation techniques before recourse to real traffic is possible. Once development is sufficiently far advanced, the simulation will be removed and an equivalent interface will be made to a real junction for field tests. (A)
Abstract