Forecasting the Short-Term Traffic Flow Based on Wavelet Transform and Support Vector Regression

A novel hybrid method based on wavelet transform and support vector regression was presented to forecast the short-term traffic speed. The method of multi-resolution was applied to analyse stochastic signal in the time series of short term traffic speed. Considering the non-linear feature in the time series, we used different support vector machines, i.e., e- support vector machine, ν-support vector machine and least square support vector machine, were applied respectively to the details and approximation from the wavelet transform to establish the hybrid forecasting model. The final prediction result of the hybrid model was obtained by integrating the forecasting results from the details and approximation. The proposed model was realized in MATLAB. Compared with some common SVR models, the performance and prediction results in our experiment demonstrated that the proposed method fitted better to the original data set obtained from real-world traffic speed data, meanwhile it had lower time cost and the predictive accuracy increased visibly.