In order to predict bus arrival time at each stop with complex real-time traffic factors changing on road, this paper proposed a dynamic arrival time prediction model for buses, using a combination of real-time taxi and bus datasets, with a kernel algorithm of statistical methods or neural networks to model the properties of dynamic road circumstances. This is a "rolling" prediction method, which uses the prediction result of previous step as the input features of the next prediction, so as to predict all the bus arrival time of future stops. The result shows that the proposed dynamic model is applicable for bus arrival time prediction, where datasets with taxi are more accurate than that without it. The possible bus bunching can be captured by checking the predicted time headway between buses from the same bus line, and long-time delay in terminals can also be predicted and eliminated by rescheduling departure time of targeted buses.

CICTP