Abstract: To explore the antecedence and accuracy of predicting soft soil roadbed settlement this study analyzed selected typical sections taking into account field relationships, engineering geological surveys, engineering measurements, and roadbed settlement tests. A combined optimization method for predicting the settling of a soft soil subgrade was proposed. First, three kinds of curve fitting methods were applied. The optimal curve was determined using identification methods for two different evaluation criteria, the minimum mean square and maximum correlation methods. Then, the FLAC3D numerical simulation method was used to predict the settling of a soft soil roadbed. The optimal fitting curve method, FLAC3D numerical simulation method, and measured data were quantitatively analyzed and combined to predict the settlement process. Prior to settling, the laboratory test data were used in the FLAC3D numerical simulation method, and the modeled values were 15% to 20% greater than the observed values. This is considered safe, and the method can be used to predict excessive settlement in early construction processes. When settling occurred, the optimal curve fitting method predicted the roadbed settlement more accurately using the measured engineering data, and the error between the final settlement and measured did not exceed 3%. Compared with the roadbed settlement after reinforcement, the joint method was more accurate; furthermore, the accuracy rate was 5% to 10% higher for the soft soil subgrade with less disturbance.