Abstract: Aiming at the problem that traditional deformation monitoring and prediction are difficult to achieve large-scale monitoring and accurate prediction, a method combining PS-InSAR technology and multi-variable long Short term memory (M-LSTM) neural network is proposed to monitor and predict the frost heave deformation of high railway foundation. Firstly, PS-InSAR technology is used to obtain the spatial distribution characteristics of subgrade frost heave. Then, Pearson correlation coefficient method is used to optimize three kinds of frost heave induced factors, and the obtained data are preprocessed to compose the training data. Finally, LSTM is introduced to construct an intelligent and multivariable frost heave prediction model to accurately predict the frost heave deformation trend of subgrade. The results show that PS-InSAR technology is reliable in large-scale deformation monitoring. The prediction accuracy of M-LSTM model is higher than that of the traditional neural network model, and the mean determination coefficient (R²), mean absolute error (MAE) and mean root mean square error (RMSE) are 0.973,0.024 mm and 0.035 mm, respectively. It shows that M-LSTM model has good application value in frost heave deformation prediction of high railway foundation, and also provides a new idea for frost heave deformation prediction of subgrade.

Key words: PS-InSAR, multivariate LSTM model, high railway roadbed frost heaving, deformation prediction