| The controlling and management of intectlence traffic is based on the real-time and accurate forecasting of short-term traffic flow. A great many of forecasting approaches for short-term traffic flow are advanced. But the precision of these forecasting approaches are dissatisfied due to the influence of all the interference don't be considered fully or the combined action of interference is transacted by single method.So the characteristic of short-term traffic flow is analyzed first by method named Principal Component Analysis and Fractal Theory in the paper. We could conclude that the short-term traffic flow possessed minimal predictable cycle is a set of chaotic time serial containing much interference, and the different component in short-term traffic flow owned different characteristic has different influence for the forecasting results. So the improved Mallat algorithm is advanced in the paper to complete information separation, which is a method used for short-term traffic flow wavelet decomposition and single-branch reconstruction, technically. And the short-term traffic flow time serial researched in the paper could be separated into low-frequency determine information, high-frequency chaotic information and high-frequency interference information. Then a double-layer wavelet network is established to forecast the information separated: the first layer of wavelet network is marked as WNN-1, which is used to forecast the low-frequency determine information and high-frequency interference information; the second layer of wavelet network is marked as WNN-2 , which is used to forecast the high-frequency chaotic information. In the end, we could get the forecasting results of short-term traffic flow by superposing the forecasting values of separated information. And the results contain the whole original signals.The results show that the method advanced in the paper is better than others introduced in the paper due to its higher precision and faster calculation speed.
|
PDF Full Text Request