Stochastic variations in traffic conditions and ridership often have a negative impact in transit operations resulting in the deterioration of schedule/headway adherence and lengthening of passenger wait times. Providing accurate information on transit vehicle arrival times is critical to reduce the negative impacts on transit users. In this study, models for dynamically predicting transit arrival times in urban settings are developed, including a basic model, a Kalman filtering model, link-based and stop-based artificial neural networks (ANNs) and Neural/Dynamic (ND) models. The reliability of these models is assessed by enhancing the microscopic simulation program CORSIM which can calculate bus dwell and passenger wait times based on time-dependent passenger demands and vehicle inter-departure times (headways) at stops.

The proposed prediction models are integrated with the enhanced CORSIM individually to predict bus arrival times while simulating the operations of a bus transit route in New Jersey. The reliability analysis of prediction results demonstrates that ANNs are superior to the basic and Kalman filtering models. The stop-based ANN generally predicts more accurately than the link-based ANN. By integrating an ANN (either link-based or stop-based) with the Kalman filtering algorithm, two ND models (NDL and NDS) are developed to decrease prediction error. The results show that the performance of the ND models is fairly close. The NDS model performs better than the NDL model when stop-spacing is relatively long and the number of intersections between a pair of stops is relatively large.

In the study, an application of the proposed prediction models to a real-time headway control model is also explored and experimented through simulating a high frequency light rail transit route. The results show that with the accurate prediction of vehicle arrival information from the proposed models, the regularity of headways between any pair of consecutive operating vehicles is improved, while the average passenger wait times at stops are reduced significantly.