Abstract: Special geotechnical tunnels,such as mudstone,silty clay,and crushed stone soil,have large deformation amounts,fast deformation rates,and complex influencing factors.Based on construction monitoring measurement data,a WMIC-LSTM model is established using weighted maximum mutual information coefficient (WMIC) and long short-term memory network (LSTM) to comprehensively consider the combined effects of different influencing factors for deep learning and predict the trend changes of tunnel rock deformation.The results show that the deformation of surrounding rock in special rock and soil tunnels has the greatest correlation with groundwater conditions (WMIC=0.21),and the selection of initial input step has a significant impact on the model prediction results.Compared with traditional LSTM models,WMIC-LSTM has smaller prediction errors for tunnel deformation curves with different deformation trends in long-term series models,with the optimal mean absolute error (MAE) reaching 0.079.The deep learning simulation prediction based on recurrent neural network algorithm meets the expected results and can provide technical means and early warning basis for intelligent monitoring and safe construction of tunnels.