Forecasting multiple-period freeway link travel times using neural networks with expanded input nodes

The ability to forecast multiple period link travel times is a necessary component of real-time Route Guidance Systems (RGS). This paper examines the use of artificial neural networks (ANN) that incorporate expanded input nodes for this problem. Recent link travel time information is used as the primary input and this information is highly nonlinear which is problematic for conventional ANN. A sinusoidal transformation technique is employed for pre-mapping the input feature (i.e. the recent link travel times) in order to increase the separability of the function approximated and hence improve the prediction capabilities of the ANN. Actual link travel times from Houston, Texas that were collected as part of the automatic vehicle identification system of the Houston Transtar system were used as a test bed. The results of the best ANN with expanded input nodes were compared to a conventional ANN, a modular ANN, and other existing link travel tune forecasting models including a Kalman filtering technique, an exponential smoothing model, a historical profile, and a real-time profile. It was found that the proposed expanded input neural networks outperformed a conventional ANN and other travel time forecasting models, and gave similar results to those of a modular ANN.