In order to study the applicability of the optimal weight combination prediction model in the settlement prediction of highway soft foundation in the seasonal frozen area, the layered settlement meter was used to monitor the settlement in a typical section of a first grade highway. The change of subgrade settlement was reflected by the movement of the magnetic ring on the settlement tube. The original nonlinear data were processed by cubic spline interpolation according to the measured settlement data of one year. The grey GM (1,1) and the conic model were combined based on the optimal weight method to predict the long-term settlement of the foundation. Afterwards, the predicted results were compared with the measured data to explore the applicability of the model. Meanwhile, selected a reasonable time interval suitable for subgrade monitoring in alpine regions based on the climatic characteristics of the seasonal frozen region. The research shows that the prediction effect of the optimal weight combination model is closely related to the number of samples, thus properly increasing the number of samples establishing the model can significantly improve the prediction accuracy. Under the consideration of regional climatic conditions, properly reducing the number of samples in winter and months with slow settlement rate and increasing the number of samples in the initial stage of settlement and before and after freezing and thawing can effectively improve the prediction effect of the model. The optimal weight combination forecasting model in this study can better predict the long-term settlement of highway in seasonal frozen area and provide reference for the settlement prediction of similar projects.