Study On Route Travel Time Prediction Based On Radial Basis Function (RBF) Neural Network

With the rapid development of intelligent transportation system, the traffic flow guidance has become the important mode of the modern traffic management system in the 21th century. How to predict the route travel time accurately is the key technology of a good traffic flow guidance system, and which is also one of the basic aspects of the intelligent transportation information system, so it is significant to study the methods of forecasting the accurate travel time in transportation engineering and control.It is rather difficult to predict the travel time using the traditional methods because of the complexity of the transportation networks. For example, the adaptability of the Kalman filtering theory is not good enough to do it, the BP neural network can predict the travel time and reflect its trend but the precision can not match with the demand of forecast performance.As a new modeling method, Radial Basis Function (RBF) neural network can learn regulations form history data, auto-cluster and organize network structure according to given issue, it can overcome the shortcoming of partial minimum point problem which occurs in the BP neural network. RBF network is wildly used for its excellent performance in many fields such as system recognition and dynamic forecasting and so on, so we put forward a real-time travel time prediction model based on the RBF neural network in this paper.This paper introduces several existing methods of travel time prediction and discusses their advantage and disadvantage, brings forward a new scheme of obtaining the road mean velocity by means of the double inductance loops based on the vehicle identification after summarizing several traditional detecting methods of traffic parameters. Then the theory and implement of RBF network are expatiated and the new method of RBF neural network is applied to predict travel time is narrated in detail. Finally, the paper compares the forecast performance of RBF neural network with that of kalman filtering theory and BP neural network via a great deal of simulation experiment. The experiment results prove that the RBF neural network can predict the travel time in real time well and the adaptability and accuracy of RBF neural network are better than that of Kalman filtering and BP neural network. Even if the travel time can't be detected with loop detector the RBF can get acceptable forecasting performance. So RBF neural network is appropriate, with respect to accuracy and speed, for use in real applications.