Li Xin1,2,3, Wei Guanjun1,2,3, Zhang Delong1,2,3
Funding Project: National Natural Science Foundation (41964008)
Keywords: PS-InSAR, Multivariable LSTM Model, Railway Subgrade Frost Heave, Deformation Prediction
Abstract : In response to the challenges of traditional deformation monitoring and prediction in achieving large-scale monitoring and precise forecasting, this paper proposes a method that combines PS-InSAR technology and Multivariable Long Short-Term Memory (M-LSTM) neural network to monitor and predict railway subgrade frost heave deformation. First, this method utilizes PS-InSAR technology to obtain the spatial distribution characteristics of frost heave; then, the Pearson correlation coefficient method is used to optimize three inducing factors of frost heave, and the resulting data is preprocessed to form training data; finally, LSTM is introduced to construct an intelligent, multivariable frost heave prediction model to accurately predict the trend of subgrade frost heave deformation. The research results indicate that PS-InSAR technology is reliable for large-scale deformation monitoring, and the prediction accuracy of the M-LSTM model is higher than that of traditional neural network models, with average determination coefficient (R2), mean absolute error (MAE), and mean root mean square error (RMSE) of 0.973, 0.024 mm, and 0.035 mm respectively, demonstrating the significant application value of the M-LSTM model in predicting railway subgrade frost heave deformation and providing new insights for frost heave deformation prediction.