In order to avoid the influence of random factors on the single prediction model, a model and method which are more in line with the settlement law of soft soil roadbed are established to improve the accuracy and reliability of the prediction model. Three prediction models, Gompertz model, Logistic model and Grey Verhulst model, were established with the original settlement data, cubic spline interpolation and piecewise cubic Hermite interpolation data as sample values, the prediction value of 3 cases of each model is compared with the measured value, and the superior method of sample processing for each model is selected. Taking the dominant method of each model as a single model, a combined model based on IOWHA operator is established. The combined model solves the optimal weight coefficient according to the forecast precision of the single model at each time. The results show that the Gompertz model and Logistic model are more effective when the data at the time when the sedimentation rate changes obviously as the last set of samples, the Grey Verhulst model is more accurate when the original settlement data is taken as the sample value. GA is more reliable in solving the weight coefficient of the combined forecasting model based on IOWHA operator than the nonlinear optimization toolbox of MATLAB, and the forecasting precision of the combined forecasting model is higher.