|
Freeway incidents is unpredictable , as quickly and accurately whether the perception road incidents, incident location and extent of damage insurance for timely relief, reduce losses, excluding events on the impact of highway capacity , highway incidents detection algorithm research has become the field of intelligent transportation systems research focus , it is to improve highway traffic management efficiency, improve traffic management effectiveness has important significance. From the characteristics of freeway traffic flow , according to the basic principles of event detection , in the analysis of the advantages and disadvantages classic freeway incident detection algorithm based on support vector machine based (Support Vector Machine, SVM) Freeway incident detection algorithm research . In introducing the basic principles of support vector machines , while focusing on the optimal classification surface , linearly separable SVM, of non-linear SVM, as well as non-linear separable SVM classification commonly used kernel function . Given event detection algorithm based on SVM work steps , design algorithms input vector . And adopt California I-880 database, designing four experiments , in each experiment were designed based on the linear non-separable SVM, Gauss kernel function , hyperbolic tangent kernel event detection algorithm to verify the effectiveness of the algorithm and portability. Different SVM models , different kernel functions and their parameters are affecting algorithm performance indicators , this paper Associate Professor Lin Zhiren Libsvm toolbox parameter optimization module penalty parameter for each algorithm to optimize the parameters C and nuclear options . Calculated for each model , and the optimal kernel function parameters, the above-mentioned four simulation experiments and compared with California algorithm , the results show that : for different experiments , select the appropriate SVM model and kernel function , get over California algorithm better performance. The algorithm freeway incident detection algorithm for improving the performance index with a reference value .
|