Study On Transportation Information Fusion Algorithm And Application Based On Improved SVM

At present, a great deal of researches have been carried out in ITS field in China, plenty of essential traffic data have been acquired through traffic management systems. However, those data were not put into use fully, with lower utilization rate. Therefore, how to take full use of traffic monitoring information to meet the demanding needs of transport, to strengthen traffic control and management, to optimize road network is a key problem urgent to resolve.In this paper, on the background of National 863 Project "Information Interoperability Technology for Distributed Transportation System" (2007AA11E226), based on the analysis of characteristics of traffic data, an information fusion algorithm based on improved SVM is brought out. The main research work are as follows:1)Centering on the support project, on the basis of the in-depth investigations on application of freeway information management systems in Jiangxi Province, Shanxi Province, Beijing etc, the collected data from local systems was studied, and the intrinsic characteristics of traffic monitoring data were summed up.2)On account of actual research project and shortages of the traditional SVM, an information fusion algorithm based on improved SVM-Decision Tree Method-Support Vector Mechines(DTM-SVM), is brought out. With "one-to-many" approach, the algorithm overcome the limitations of traditional SVM effectively which is only suitable for two-classification problems, and figures out multi-classification problems, meeting wider applications.3)Centering on DTM-SVM algorithm, the hierarchical structure model of DTM-SVM was established. And the three important links of fusion process- SVM training, training results evaluation and SVM testing, were elaborated.4)Relying on research background, on the basis of the establishment of evaluation index system of freeway traffic state recognition, the DTM-SVM was applied to solve the problems of freeway traffic state recognition. Results of the application show that the algorithm overcome the limitations of traditional SVM effectively which is only suitable for two-classification problems, and be applied to wider fields.