In order to analyze the irregular height deformation of the Qinghai-Tibet railway subgrade, by establishing an elevation-time response model, based on the Box-Jenkins modeling method, determine the order of the time series model, select a suitable time series model according to the AIC criterion, and finally give a batch forecast. The algorithm steps for measuring the elevation of subgrade points. The problem of the time-deformation law of the elevation of the Qinghai-Tibet railway subgrade is studied. The results show that: Taking the monthly elevation data of the left subgrade at K1425+050 of the Qinghai-Tibet Railway from 2010 to 2018 as an example, an ARMA (2,1,1) model was established, and the 2019 data was used as the validation set. The model passed the model The adaptability test proves the validity and accuracy of the model; it summarizes the possible major deformations in the predicted values of the measuring points along the Qinghai-Tibet Railway to December 2023 and the large differences in the elevation difference between the left and right sides of the measuring point. 10 dangerous points; around the measuring point K1476+600, there are obvious long-distance differences on both sides of the roadbed. It can be seen that this model can accurately predict the elevation change of the Qinghai-Tibet Railway subgrade, which has certain reference significance for engineering maintenance and repair.