A method of traffic flow forecast and management

As an important aspect in transportation research, the forecast of traffic state serves as the basis for reasonable traffic control and management. Original prediction models in actual appliance don't consider influencing factors comprehensively when dealing with complex and changeable traffic state. Based on the observation of daily similarity of expressway traffic flow series, the paper recognizes weather and date attribute as classification vectors and designs a new time series clustering algorithm through calculating representatives in all kinds of clusters and building correlative relations between observing time series and representatives. In the end traffic flow under different time scales on both weekday and weekend are forecasted. The experiment result shows that the time series clustering algorithm and forecast model make good use of historical data and the accuracy is quite satisfied under 30-minute time scale with the Mean-Absolute-Relative-Error 3.34 percent.