To ensure the safety of train operation in strong wind environment, a combined prediction model based on VMD (variational mode decomposition), LSTM (long short-term memory neural network) and RIME (frost and ice optimization algorithm) was proposed. Firstly, VMD was used to decompose the original wind speed time series on a multi-scale to obtain modal components with different frequency characteristics. Then, RIME algorithm was used to optimize the hyperparameters of LSTM model to improve the performance of the model. Finally, the experimental verification was carried out based on the wind speed data of several groups along the railway. The results were compared with those predicted by LSTM, EMD-LSTM, VMD-LSTM and VMD-RIME-LSTM models, and a number of error evaluation indexes were obtained. The VMD-RIME-LSTM combined model proposed in this study shows significant advantages in forecasting accuracy and stability. The research results can provide theoretical basis and technical support for wind speed monitoring and early warning along railway lines, and have important practical significance for improving railway running safety.