Travel Time Prediction by Advanced Neural Network

Abstract

The Advanced Traffic Management System of San Antonio, Texas, called TransGuide System uses a sensor system installed in 26 miles of highway to feed data to a high speed computer network for analysis. The portions of interstates involved were generally confined to central city areas and did not reach the first outer loop that surrounds the inner city. The objective of this paper is to build a real-time travel time prediction model for the freeway network of San Antonio based on the information collected by the loop sensor and GPS systems. The travel time prediction of the model could be the basis of later traffic management systems and also used by the traveler information systems. The robustness and accuracy of the model is a very important feature because traffic management systems depend on driver acceptance and compliance to be effective. This paper examines first the use of Modular Neural Networks (MNN) to forecast multiple-periods of traffic engineering features, such as speed, occupancy and volume, and then determines the expected travel times based on these predicted values, using currently applied methods. Secondly, the multiple-periods travel times are predicted directly from the loop data with an MNN. The models are tested and trained on actual travel times from San Antonio, collected by GPS data system. Then the results of the two models are compared to each other and to the results of standard travel time prediction models.