Study On Traffic Flow Processing And Predicting Model

This paper takes traffic flow prediction of crossroad as research subject, presents a method that combine wavelet analyze with least square method to find out the abnormal data and repair it; then using an improvement BP algorithm to build the model of traffic flow prediction.The main study work of this paper as follows:â‘ Collecting traffic flow data to be as time series and decomposing it with two layer wavelet analyze, take the reconstruction series of low component which decomposed as the original traffic flow series reference series; then using least square method to find out the abnormal data and repair it.â‘¡In research of traffic flow predict model, using improvement BP algorithm which take change step method to offset the defects of traditional BP algorithm which is easy to trap in local minimal solution and have slow convergence when the times of training increase.â‘¢Taking a practical crossroad for example, this paper designs two models of traffic flow prediction, one uses historical traffic flow data itself of predicted section as samples of training, the other combines the historical traffic flow data itself with others traffic stream's as samples to training the net.A great deal of experiments on the crossroad of Chongqing have be done, the results show that using the method of combining wavelet analyze with least square method can gain a high precision; in the traffic flow prediction, the nets of using itself historical traffic flow data to train is better than combining others traffic stream's.