The Research Of Traffic Flow Prediction Based On Wavelets Packet Analysis And SVM

According with the development of economic, expanding of cities, the developing cities are facing the problem of traffic congestion urgently. Nowadays, the problem of traffic jams and congestion is a direct acting factor for quality of life. The current traffic command scheduling method almost depends on the time, but not on actual situation of the road traffic flow status. The traffic flow is high uncertain and complicated composition. Traditional traffic flow predictive methods are used to deal with traffic data directly in time domain, whose anti-interference ability is poor and data approximation performance is bad. On the analysis of traffic flow characteristic basis, this paper considers synthetically the advantages and disadvantages of existing prediction method, so that this paper adopts wavelet packet analysis and support vector regression machine for traffic flow forecast, to enhance practicability and accuracy for the prediction of traffic flow.First of all, the original traffic flow data should be decomposed by wavelet packet analysis theory, and the original data is decomposed into four frequency band traffic flow characteristic data according to the different frequency of energy. Then, we reconstruct these four frequency characteristic data and analyze the characteristic data using ordinary decompose tree and the optimal decomposition tree to get four-dimensional traffic flow characteristic data that have the same length with original data. Then, the processed traffic data are divided into the training sample data and test data.Finally, the forecast of the reconstructed traffic flow characteristic data should be made using support vector regression machine. This paper carries out traffic flow forecast separately in normal weekday and weekend. Kernel functions, punish factor etc can be sure after training the support vector regression machine, the data of traffic flow prediction as outputs could be got after inputting the rest of the traffic flow characteristic data into the trained support vector regression machine model. Meanwhile, we test the trained support vector machine (SVM) model using test data as inputs, and do the error analysis for prediction. We realize that it is more accurate to use the prediction methods in this paper comparing with the traditional neural network and time series prediction algorithm. This paper provides a new method and new ideas for the research of traffic flow prediction.