A GMDH-based traffic flow forecasting model

Traffic flow forecasting, the core element of intelligent transportation system, plays an important role in traffic information services and traffic guidance. Since neural network prediction needs plenty of training samples, it cannot guarantee the real-timeness of traffic flow forecasting. In this paper, a GMDH network was constructed by self-organization, and the network was applied to traffic flow forecasting, hence a GMDH-based traffic flow forecasting model. Owing to the fact that a GMDH network is trained without much prior knowledge, and the network structure is characterized by self-organization and globally-optimal selection, a short-time prediction model by the network has a good performance. The model was simulated with MATLAB, and the simulative results indicated that the model was the right tool for data-fitting. The average relative error of prediction was only 3.35%, and the model was valid.