Travel time prediction with non-linear time series

This paper focuses on a specific application of Advanced Traveler Information Systems (ATIS), which is the posting of travel time predictions on Changeable Message Signs (CMS), The objective is to influence travelers' route choice. One of the statistical techniques that has strong potential for on-line implementation is the non-linear time series with multifractal analysis. In this paper, a new approach for predicting travel times is developed and tested with travel time data for five incident free clays with the goal of predicting recurrent congestion.. The prediction errors were found to have the greatest magnitude at the temporal boundaries of congestion. Refining the prediction algorithm through the smoothing of the input data and setting a threshold on the minimum speed predictions improved results by abating the influence of the temporal boundaries of congestion. Thus, the new approach produced reasonable errors for short-term (5-minute) travel time predictions.