Freeway Traffic Flow Prediction Based On BP Neural Network

Traffic Flow Guidance System (TFGS) take an important part in Intelligent Transportation System (ITS). Traffic flow prediction is the core problem in the TFGS. How to predict traffic flow amount online is the key solution to the TFGS.This article applied the neural network to the TFGS. The traffic flow is changeable and nonlinear system. Firstly makes a comprehensive survey of freeways' traffic flow models. Based on the analysis of some macroscopic dynamic traffic flow models frequently used in traffic control and simulation, then it put forwards an improved traffic flow model, which can describe traffic flow's real behavior accurately.Secondly used the BP neural network and its improved algorithm, the RBF neural network establish the traffic flow forecast model. Through the concrete example, use mathematical instrument MATLAB6.5 and its neural network toolbox, has produced the corresponding simulation curve and the forecast result, and has carried on the comparison and the analysis with the primary data.The simulation results show that the above neural networks are relatively ideal tools of traffic flow prediction. But these neural networks are different from each other in terms of time cost and accuracy of the simulation process. BP neural network is not satisfying in terms of prediction time and accuracy. By adopting improved arithmetic, the accuracy can be made higher and prediction time can be made shorter. RBF neural network is with shorter prediction time but higher dispersion degree and lower prediction accuracy. The conclusions can be made as below. First, BP model is more advanced than RBF model in prediction accuracy. Second RBF model is more advanced than BP model in terms of speed and sample points approaching performance. Third BP model is more inclined to vibrating while RBF moder is more inclined to stabilizing. Fourth defer prediction at certain point is likely to happen in both of the two models. In a word, the two models are comparatively good methods for short term traffic flow prediction. Actual traffic leading requirements can be met by the two model's prediction accuracy.