Study on a New Traffic Flow Forecasting Method

Traffic flow forecasting is a popular research topic of intelligent transportation systems (ITS). Some forecasting models have been developed, but these methods´ precision usually can´t meet with practical requirement. The Traffic models of different time sections in a day have diversities. The forecasting precision could be improved if the models are built on different time section. A serial clustering method based on extended entropy information bottleneck theory is proposed. It is used to partition a day into different time section according to history traffic flow data. Support vector machines (SVM) is used to forecast traffic flow. Bayesian inference is used to fix the SVMpsilas kernel parameters to improve the regression precision. The efficiency of the method is illustrated through analyzing the traffic data of Jinan urban transportation.