Expressway Traffic Volume Forecasting Based On Support Vector Machine

Expressway is not only an important symbol of modern transportation, but also a vane of the national economic development level. The Chinese highway has developed quickly since 1988.Eespecially after "National 15 Plan", the government invest more money in expressway construction. At present, the expressway mileage is second only to the United States and will soon become the world's longest state expressway mileage.With the rapid development of the expressway highway management is also very important. According to the government plan, Chongqing municipal government will build on second-circle highway and will undo Chongqing expressway toll of all inner before the end of 2009. Then they will build new expressway toll near the city of rays direction. The main loop freeway will be buyed back by the government as the city road. The expressway between inner rays will be leased by the government. The expressway will toll in year. So in the process of the buyback and leasing, determining the future of highway traffic accurately, reasonably will become the key to determine the value of highway.Under this background, this paper research in the highway traffic data. Then it was accurately and reasonably predicted. Firstly, this article reviews the existing prediction theory and method and focus on the characteristics and the applicable scope of the method. Secondly it point out characteristics of the highway traffic flow in China. Due to the short history in the national highway construction and imperfect construction of information system, the traffic data is lesser. Also the size of the highway traffic is relative with and economic factors of a region, such as economic development and population.According to the forward work, this paper chooses Support Vector Machine (SVM) to predict the annual expressway traffic volume. Support Vector Machine is a new Machine learning algorithm based on statistical learning theory which putted forward by Vapnik. The reality of (SVM) is structural risk minimization principle. Compared with the traditional machine learning method, SVM in solving the small sample, high dimension and nonlinear problem has obvious advantages. Just generalization performance, SVM has good generalization. In statistical sense less support vector is corresponding good generalization ability. From the simple characteristics of support vector machine, support vector machine is suitable for the annual volume of the prediction models. This paper tests the accuracy and consistency. In the application of the support vector machine (SVM) method, kernel function of support vector machine and parameters selection is mainly studied. Through the test samples, it choose more optimal kernel functions. Particle Swarm Optimization algorithm is used to optimize the parameters.