Abstract: To reduce the impact of communication delay on intercity railway, this paper utilized the historical data of communication delay time of train-ground GSM-R accumulated by communication control server(CCS) to form the delay time estimation algorithm framework, used recurrent neural network and standardized deep learning technology, combined with wireless transmission packet size and delay time, to learn and training the data and model. Experimental results showed that this method could effectively predict the delay time of wireless transmission, and solve the problem of low accuracy caused by statistical analysis, provide the basis for accurately judging the effectiveness of train-ground communication message.