Short-term Traffic Flow Mixed Forecast Methods Study

The prediction methods of the short-term traffic flow were presented. Real-time freeway traffic flow is used to study the methods which are also suitable for the city traffic flow too. First the important meaning of the study and the current prediction methods was introduced and the character of short-term flow was analyzed. Based on the non-linear, time-varying, complexity of the flow, traditional methods is not suitable for the prediction, so two new methods are proposed in the thesis.Method one: according to the good time-frequency analysis of wavelet transformation, decompose the traffic flow into different scales and reconstruct each component single. High frequency and low frequency components could be gotten, which is simple and stable relatively. For each component different method can be adopted. From the autocorrelation and partial correlation function of the component, ARMA model is decided to predicting the high frequency and GRNN for the low frequency; then add all the new components and the result is the integrated prediction one.Method two: based on the fundamental chaos theory, calculate the characteristic parameter of the traffic flow; delay timeτis calculated by the autocorrelation function method and embedded dimension is calculated by the G-P algorithm. Then reconstruct the space, a new space which is similar to the original one has been gotten. Combine the new phase space components with the neural network; take it as the input of the RBF network, the output of the network namely is the final prediction results.The Mat lab simulation graph and the performance index parameter present that the two proposed methods have gotten high precision and is good enough for the real-time controlling and guiding of the traffic flow.