Parallelized Traffic Prediction with a Nonlinear Model

For online traffic control at traffic control centers there is a need for predictions of traffic flow over a short horizon to evaluate the impact of different scenario for the purpose of on-line scenarios selection. Fast computation is needed for the online decision support system. However, the common approach of a parallel simulation has several drawbacks. In this paper, a novel approach by an asynchronous iterative algorithm is presented to predict the traffic flow in a large-scale traffic network in a parallel and distributed way with a reasonably small computation time. By a proper selection of the number of distributed subsystems one can reduce the time complexity of the prediction to a value satisfactory for traffic control centers. The experience with the computations is that after two or three iterations the obtained predictions for the prediction horizon are quite satisfactory for practical use. A theoretical analysis of the convergence problem is in progress.