Study On Urban Traffic Prediction And Monitoring Technology Based On Simplified Road Network Model

Quickly and accurately obtaining real-time road condition and predicting futureroad condition can help discover existing traffic congestion in time and foresee possiblecongestion in advance. It is not only important for the traffic management department tocontrol traffic effectively, but also provides a basis for travelers to choose travel route;it’s the foundation of dynamic traffic management. Currently almost all the trafficmonitoring study does not distinguish congestion between intersections with links. Thispaper adopts a new simplified network model and studies on the urban road trafficparameters prediction and traffic monitoring, the main research contents and results areas follows:Firstly, a single-step prediction model of traffic parameters combined SupportVector Machine and Kalman Filter is proposed.By analyzing the advantages and disadvantages of the existing predictionalgorithms, the paper takes advantages of the "historical traffic pattern" contained inSupport Vector Machine method and “real-time traffic trends” contained in KalmanFilter method, and proposes a combination forecasting based on the two predictionmethods. The combination way is simple and intuitive: the prediction way taken in nexttime period is determined by the double standards of error square sum and correlationcoefficient between measured values and predictive values in the past several timeperiods, which is proved to be universal and stable.Secondly, a multi-step ahead prediction method of traffic parameters based onSupport Vector Machine and History Similar Sequence is proposed.In the case of short prediction steps, multi-step iterative prediction based on supportvector machine is adopted. With the increasing of prediction steps, in order to solve theproblem of error accumulation in the multi-step iterative prediction, the History SimilarValue is introduced, that is, the same trend of traffic data sequence as the predictive dateneed to be found from the historical traffic database. It reduces the accumulative errorcaused by iteration to some extent, and slows down the growth rate of the multi-stepprediction error.Thirdly, a calculation method of urban traffic based on simplified network model isdesigned.Roundabouts and overpasses are abstracted nodes and a road link has more than one weight in simplified network model, so the traffic identification ways of ordinaryroads, roundabouts and overpasses are designed respectively.Finally, the proposed single-step and multi-step prediction algorithms of trafficparameters are verified and analyzed by experiments; meanwhile, the traffic monitoringand prediction system based on simplified network model is implemented.The proposed single-step and multi-step prediction methods are verified by trafficflow data from University of Minnesota Duluth and compared with the existingmethods. The experiments suggest that the combination model of single-step predictionoutperforms each single prediction model. It plays the advantages of both predictionmethods. Multi-step ahead prediction can effectively predict the traffic flow within anhour in the future, and the relative error is maintained at less than10%during peakperiods. At last, by utilizing the simulated traffic data sets, the traffic monitoring andprediction software is implemented on a regional map of Chongqing, which can displayboth the real-time and predictive traffic for users to view.