In the process of highway tunnel construction，the stability of surrounding rock has a great impact on tunnel construction.Therefore，the monitoring measurement and accurate prediction of surrounding rock deformation of highway tunnels are the keys to ensuring the safety of tunnel construction.In view of the low prediction accuracy and poor generalization ability of tunnel surrounding rock deformation，this paper proposed a Bayesian （Bayes ）-based method to optimize the long-term and short-term memory （LSTM ） network.The method first preprocessed the original monitoring data of crown settlement and peripheral convergence，then constructed the initial LSTM model of crown settlement and peripheral convergence of highway tunnels，and used the super parameters in the Bayes optimization model to obtain the prediction results.The model was used to predict the crown settlement and peripheral convergence of a highway tunnel，and the prediction results were compared with convolutional neural network （CNN ） and support vector regression （SVR ） using root mean square error as the evaluation index.When the crown settlement was predicted，the average prediction accuracy of the Bayes-LSTM model was 1.0 and 1.26 higher than that of the CNN and SVR models，respectively.When peripheral convergence was predicted，the average accuracy of the Bayes-LSTM model was 0.3 and 0.32 higher than that of CNN and SVR，respectively.The results show that the Bayes-LSTM model has higher prediction accuracy，and it can judge and choose the historical information in the process of model training，which greatly improves the efficiency of time series data processing.The model provides a new idea for the prediction of surrounding rock deformation of highway tunnels.