The service environment in which the lining structure of tunnels in high cold regions was usually harsh and susceptible to the coupling effects of various extreme factors, leading to different forms of cracks in the lining concrete and posing significant safety hazards to the entire tunnel structure. Therefore, monitoring and predicting the cracking behavior of tunnel lining structures in high cold regions has important engineering significance, so, a multi-factor coupling prediction model for tunnel lining strain in cold regions based on deep learning is proposed. Firstly, a CNN-Bilstm-Transformer (CBLT) multi-factor coupling prediction model was developed based on the LSTM model, using a high cold region tunnel of Xinjiang as a supporting project. Then, based on a large amount of on-site monitoring data, CBLT was trained and tested to explore the influence of different input features on the model’s predictive performance, and the predictive performance between CBLT and different models was compared and analyzed. The research results indicate that: With the improvement of input features, the predictive performance of CBLT gradually improves; the average RMSE, MAE and MAPE on the test set were only 2.61 με, 2.22 με and 1.24%, R² can reach 0.815; the predictive performance of CBLT is better than RNN, LSTM, CNN-Bilstm and Bilstm-Transformer. The application of CBLT in predicting lining cracking behavior in practical engineering could achieve high accuracy. The research results can provide scientific guidance for the analysis of cracking behavior of tunnel lining in high cold regions and the prevention of frost damage.