We apply the wavelet transform to forecast the traffi c load in this thesis. The traffi c load time series are decomposed into several detail signal and approximation signal by the wavelet decomposition. We use AR model or the cosine approximation model to fit the detail signal

and polynomial fitting and AR model to fi t the approximation signal. Using the traffi c of gaming day as the test sequence which from Jan-2009 to Jul-2013

the fi rst 50 months of the data to model fi tting comparison

the last 5months of data to forecast. We found the fi tting correlation degree is 0.991

the average absolute error of prediction is 0.029.