Researh Of Freeway Traffic Incident Detection Based On Support Vector Machine

The rapid development of highway construction has brought us significant economic and social benefits. However, with the growing traffic demand, the freeway traffic accidents frequently occur, which seriously affect the traffic capacity and operating efficiency. Consequently, how to detect the traffic incidents rapidly so as to take measures to reduce the traffic delay effectively and ensure the safety of transport has been paid much attention. In recent years, the traffic incident automatic detection technology which aims to solve the above problems has become a central isuue in intelligent transportation field. Its performance can directly influence the freeway incident detectoin effects, therefore, it is quite essential to research on this issue.Based on the characteristics of freeway traffic flow and the basic principles of traffic incident detection, this thesis reserches on freeway traffic incident detection based on Support Vector Machine(SVM).First of all, taking the difficulties such as the traffic data sample is limited and the input traffic characteristics is too redundant into considertation, the thesis designs the freeway traffic incident detection algorithm based on single SVM which is divided into three parts: traffic data pre-processing, SVM model constructing and decition outputing. In the simulation process, the data normalization method is used to process the traffic data to boost the detection accuracy and shorten the detection time; Then the principle component analysis method is adopted to extract the traffic flow characteristics, which aims to reduce the demention of the traffic data, the computational complexity and the SVM modeling time. At last, the genetic algorithm is used to select the parameters of SVM model, which aims to boost the freeway incident detection accuracy. Secondly, based on the ensemble learning theory, the thesis analyses the Bagging and Boosting method which both belong to ensemble learning. Then both of the methods are combined with SVM to simulate the freeway traffic incident detection separately.The simulation data is extracted from I-880 traffic data set. Comparing and analysing the simulation results, the SVM ensemble method achieves better comprehensive detection performance with less modeling time and higher detection accuracy, which provides a method for design well performance for freeway incident detection.