The Study On Short Time Traffic Flow Prediction

A new method for short time traffic flow (STTF) prediction is presented in this thesis based on wavelet, fuzzy neural network and autoregressive model. First, we review the history and status on STTF prediction. Then methods of traffic flow prediction used now are analyzed. Because of the nonstationary and nonlinear properties of STTF, the methods presented until now are not good for STTF prediction. So our new method uses the Wavelets to decompose STTF data to different scale (frequency) spaces and predicts each scale space separately. For the sake of better prediction of large-scale data, we suggest to use the wavelets that have better smooth property.As we know that the hidden layer number of Neural Network (NN) has great impact on the precision in STTF prediction and there is no theory on how to chose the number of hidden layers. The structure of Fuzzy Neural Network (FNN) is decided by the fuzzy rules. Furthermore, FNN can be used to fit the smooth curves perfectly. So we use FNN to predict the large-scale data from the wavelet decomposition.Autoregressive model can predict the low-scale (high frequency random) data as well as FNN, but it is simpler than FNN. So we use AR to predict high frequency data.We perform simulations for a certain 30s STTF prediction using different approaches and compare the results. It shows that our method greatly improved the precision of STTF prediction.