Traditional bus arrival time prediction such as Artificial Neural Network (ANN) and Support Vector Machine (SVM) is based on large-scale data mining. Due to the real-time and uncertainties characteristic of traffic conditions, the method based on historical data mining cannot predict bus arrival time accurately. This paper presents a real-time prediction method based on probe bus fleet models. The probe bus fleet is established by all the buses running on the bus line. The bus line will be separated into different meta-segments because of the overlap of different bus lines and etc. The bus fleet should be established and disassembled dynamically based on the meta-segment, and the real-time data collected from probe buses could be used for next bus arrival time prediction by using the Kalman filtering technique. Experimental results show that this model provides a higher level of veracity and reliability of travel time forecasting in the case of frequently changing traffic conditions, and support real-time adjustment to obtain more accurate bus arrival time.
展开▼
