During the filling construction of the soft soil roadbed

the total settlement value could be predicted by using polynomial and time-series combined model. On the basis of measured settlement data

the polynomial prediction model could be discriminated and established by statistics; and the settlement trend value at a certain future time would be predicted. The random settlement could be gotten by random prediction model that is established by smooth and stable time series analysis method. The sum of above two prediction values is the settlement prediction value of the roadbed at some certain time. With the addition of new settlement data

the model parameters are continuously modified. The practical settlement prediction of the soft soil roadbed shows that the combined model is superior to the single polynomial model; and the absolute error value between measure and one-step prediction value is mostly below 5 cm. The more prediction step is

the greater error is.