Application of Bayesian Trained Neural Networks to Predict Stochastic Travel Times in Urban Networks

摘要

Urban travel time prediction has received much less attention than freeway prediction, because urban travel times are much more stochastic. However, urban travel forms a significant part of total travel time. In this paper, neural networks are used for urban travel time prediction as these are able to deal with noisy data. Bayesian techniques are used for training, resulting in committees with lower error and in confidence bounds. It is shown that the networks are capable of predicting the ‘trend’, obtained through de-noising, with an error in the same order of freeway predictions, and that they accurately predict confidence bounds.