In a dynamic (real-time) traffic control system, the accuracy oft he prediction of traffic characteristics such as flow, speed, and headway is one of the key factors affecting the performance of the system. Because the traffic characteristics can be described by stochastic processes, nonlinear an time-variate types of prediction models could be more adequate than linear or time-invariate prediction models. A traffic control system model for traffic flow is described, and the importance of the accuracy of the prediction model is emphasised. Then, the concept of adaptive prediction of traffic flow is introduced, and its mathematical derivation and the least-mean-squarealgorithm are described. As an experiment to validate the adaptive prediction system, a sine function is used to simulate traffic flow as input to the adaptive prediction system. Finally, the adaptive prediction system is applied to actual traffic flow data collected from a highway network. The predicted traffic flow is then compared with the real traffic flow. The performance of the model as to its dynamic response to a step function, convergence of the adaptive prediction system, and related matters are also discussed.
Abstract