Study On The Forecasting Of Short-term Traffic Flow Based On Chaos And Neural Network

The short-time traffic flow forecasting has always been a quite active but not satisfying research topic. In recently years, many academicians have regarded traffic system as a nonlinear deterministic kinetic system. Using the rules of nonlinear deterministic system to study the traffic flow shows more and more vitality. With the development of nonlinear theory and artificial intelligence, Chaos theory and neural network become cogent tools for traffic system analysis and forecasting.First, in view of many problems that still exist in BP net training, such as a slow training speed and easily getting stuck into local minima. So we've analyzed conventional BP algorithm and usual improving technique, found out the reason of those problems. We tried to propose a new algorithm based on former effort to overcome these problems in our forecasting work of the traffic flow. We mainly altered two aspects of the original algorithms: transfer function and learning rule.1. We advanced a new transfer function and revised the stride factor with the momentum factor.2. To solve the efficiency problem of the algorithm with fixed learning rate, we proposed a novel global optimization learning rule or the basis of the conjugate-gradient method, and analyzed the convergence of it.Second, we apply the algorithm to the forecasting of the short-time traffic flow at some road in Jing-zhu highway. During the application, we compare the results with other existing forecasting technique. From the comparison, we found the new algorithm has high accuracy and efficiency comparing to traditional BP algorithms.After that, this paper has carried on the chaos recognition to the short-time traffic flow and confirmed the short-time traffic flows exist chaos phenomenon. Based on the improved BP algorithm, the single variable data is reconstructed with chaos theory, and then ANN is used for prediction. Comparing the data which is not reconstructed, we get conclusions as followings: it is certain to affect the forecast accuracy of single variable data through reconstructing with chaos theory because the difference is clear between the prediction with chaos theory and without chaos theory.In the end, we drew the conclusion, the improved BP algorithm enhanced the training speed greatly, basically solved the partial minimum problem, and has provided one new ponder question method for the neural network technology. Besides, the synthesis forecast method based on the chaos neural network can adapt the accuracy and the timely request than the sole neural network method in the short-time traffic flow forecasting.