Research On The Application Of GA-SVM In Freeway Traffic Incident Detection

In recent years, the traffic accidents have occurred very often, and it has greatly affected the "safe, fast, high efficient, environmental protecting" image of freeway. How to detect fleetly, make the judgment, and take the measures timely , to reduce the traffic delay efficiently, ensure the safety on the road and decrease environmental protection, becomes a high attention. With the development of network technology, communication technology, and automatic control technology, the intelligent accidents automatic detection system has provided the efficient way of solving the above problems. While the accidents detection algorithm is the core content of automatic detection system. Its performance can affect the efficiency of accidents detection system directly, so the study on it has the very vital significance.Through analysis and summary of commonly used incident detection methods, it is found that although these methods can perform well, they are also unable to achieve a more superior performance, because of the problems such as the traffic data sample is limited, the input feature is too simple or redundant and the parameter establishment is unable to achieve the most excellent. In view of the above questions, based on the analysis of the classification principle and predominance of support vector machine, the paper designed support vector machines model parameter and feature synchronization selection method based on genetic algorithm in detail, comprehensively using the support vector machine technology which is in view of limited sample and genetic algorithm optimization technique, in order to optimize each support vector machine's model parameters and select the best model corresponding feature combination at the same time. In view of the fact that the above methods theoretically has generalization ability in allusion to limited sample and has ability to select features and optimize parameters, the paper constructs each kind of genetic- support vector machine incident detection model, using the I-880 actual data set. In analysis, a comparison and appraisal in aspect of detection performance, extensive ability and determination speed has been made to obtain the most superior model. In order to verify the necessity to select the most superior feature combination and the validity of the genetic selection method, based on grid search method, the support vector machine incident detection model has been constructed. Aimed to prove the superiority of support vector machine method, the neural network incident detection model has been constructed to make comparison with the above two models.With the limited traffic data sample, appropriate traffic input features, optimized model parameters, the genetic-support vector machine has obtained the satisfactory examination effect, which provides a method for design high performance's incident detection algorithm. Therefore this research has a practical significance.