Research On The Traffic Information Fusion With The Multi-Source Detectors

As an effective way to solve the problem of urban traffic congestion the intelligent transportation system has been unanimously approved by lots of cities.With the license plate recognition,RTMS detector and other testing equipment continues to enrich,it largely promoted the intelligent development of traffic management.However,there are some differences in the format and type of traffic flow parameters obtained by different equipment,and on the other hand a single detection equipment often exist lack of information,abnormal quality problems.This will lead to the wrong decision.Therefore,how to achieve a variety of equipment information complementary checks,and thus improve the accuracy and reliability of traffic flow parameters information is particularly important.Based on the data of traffic flow parameters of Beijing Second Ring Expressway,this paper proposes a data fusion scheme for multi-source heterogeneous traffic flow for interval real-time vehicle speed estimation.Firstly,analyze the properties of the original data and the temporal and spatial correlation of the traffic flow based on multi-source dynamic traffic flow parameter data.Secondly,by comparing the advantages and disadvantages of the fusion method in the field of transportation,the method of this paper is determined.Thirdly,according to the data quality of traffic flow parameters,this paper studies and corrects the order of time-point drift data,missing data,redundant data and error data,uses the dynamic threshold identification method and the traffic flow mechanism identification method to identify the wrong data,and choose a different method to repair the data according to the missing situation,and proposes an error data correction method based on improved Edgar interpolation,and then the Kalman filter is used to modify the data.Finally,established a fusion model of wavelet neural network fusion model and adaptive mutation particle swarm optimization wavelet neural network,and the models were evaluated by introducing the error indexes such as MAE,MAPE,RMSE,MSPE,EC and LSE.The results show that the fusion accuracy of the two methods is better than that of the traditional BP neural network,EC value is more than 90%,the degree of dispersion of the error is smaller,and validates the importance of data processing.