Hybrid model for prediction of real-time traffic flow

Abstract

Effective prediction of real-time traffic flow is important for traffic management and intelligent traffic systems. This paper proposes a hybrid model, consisting of the k-nearest neighbours (k-NN) method and the Kalman filter (KF) technique, to dynamically predict real-time traffic flow. In the model, the k-NN method predicts a baseline speed of traffic flow on the basis of historical travel data of the target road link. To reflect the dynamic evolution of traffic flow in the prediction, a KF-based algorithm that uses the latest travel data, is developed to adjust the baseline travel speed. The hybrid model is tested with global positioning system data of Foshan City, China. In the numerical test, the proposed hybrid model is compared with a single k-NN model based on the same database. The results show that the hybrid model can provide more accurate prediction and thus holds potential for use in practice.