Study On The Short-term Prediction Method Of Traffic Flow Based On State Division

Serious traffic problem appears because of urbanization in many cities. Intelligent Transportation Systems(ITS) is one effective approach to solve it. And traffic control and route guidance is very important in ITS. At the same time, the short-term prediction of real-time traffic flow is key to realize the traffic control and route guidance. People have designed many prediction models based on traditional traffic flow theory, which describes uninterrupted parameters in two-dimension plane. So above models cannot reflect the break point in actual traffic. In fact, demonstration indicates that traffic flow changes obviously in the break point. Therefore, prediction precision falls. In this paper, inspired by "three-phase traffic flow" theory, a short-term prediction method of traffic flow based on state division is studied. First, by wavelet analysis, the actual data of Beijing second-ring road is divided three different states according to "three-phase traffic flow". And the domains of traffic flow parameters (flowrate, velocity and density) of different states (free flow, synchronized flow and wide moving jam) are gained. And then, the single Neural Network prediction model and Kalman filter prediction model are designed respectively, combining with analysis on the actual traffic flow data. The prediction method based on state division is this paper's core and innovation, which can judge if traffic flow breaks dynamically according to the domain of density, in the course of prediction. And correspondingly, the next plan is determined. Thereinto, Kalman filter prediction model is used in free flow and Neural Network prediction model is used in synchronized flow. It is proven that the method based on state division can improve the precision comparing with single Neural Network model and Kalman filter model.