Dynamic signal optimisation for isolated road junctions
The operation of traffic signal has evolved from fixed-time control to traffic-responsive control. The main difference between these two methods is the kind of traffic flow data used for signal timing optimisation. Various techniques have been developed in the literature for traffic-responsive signal control at isolated junctions: non-optimising methods and optimising methods, which differ in their use of detector data and the presence of an objective function. However, the optimising methods developed in previous studies were subject to various simplifications in respect of one or more rules of operation, in ways of interpreting the detector data, specifying the objective function, and making control decisions. In this study, a systematic approach is presented to develop an optimising method for use in dynamic signal control. Methods of this kind have two main parts: a traffic model and a dynamic optimiser. They are processed in alternating order; the dynamic optimiser provides signal timing plans to the traffic model, and subsequently the traffic model provides estimates of operational performance back to the dynamic optimiser. In this way, the dynamic optimiser makes control decisions, either to extend the current stage or to terminate it. The key feature in this approach is that a plan is developed for the entire lookahead period, but only the first part (one time-step) is implemented. In order to overcome the simplicity of the vertical queueing model, the concepts of kinematics in physics are applied to develop a novel traffic model (the KCS traffic model). Hence, it can be used to interpret detector outputs, and then to estimate three distinct components of delay (detection period delay, prediction period delay and terminal cost) that are used in the objective function. A stage-based exhaustive search method is integrated with the dynamic optimiser within a rolling- horizon formulation. This kind of search method will identify the best plan at that time for the objective in hand, though further arrivals may render that sub-optimal in the future. The lookahead period proposed here is the same for all streams, but varies according to the state of the dynamic optimiser. This optimising method is developed to be used together with various traffic models and objective functions. The formulation is presented on a stream-by-stream basis, so that it is applicable to a wide range of junction configurations. The performance of this optimising method was compared with existing control methods (System D Vehicle Actuated and fixed-time) by interfacing with the microscopic traffic simulator SIGSIM. The final results were presented in terms of the mean rate of delay with associated standard error to the number of runs. The results show that such a systematic approach can provide substantial improvements in the junction control performance and gives less delay than the existing methods.