Aiming at the throughput of high-speed railway Multiple Input Multiple Output (MIMO) system has not been fully improved, an adaptive beam transmission scheme based on antenna grouping was proposed. Firstly, the train position information was predicted by the Base Station (BS), and the beamforming technology was introduced into high-speed railway environment to establish a high-speed Massive MIMO three-dimensional model. Secondly, it was verified that in BS antenna grouping situation, the throughput of a sub-beam and its corresponding number of transmit antennas satisfied nonlinear relationship and the number change of sub-beam antennas did not affect the throughput of other beams. Finally, based on the above, an adaptive beamforming scheme based on antenna grouping was used to adjust the number of beams required and the number of transmit antennas required by the sub-beams when the train runed at different locations to ensure optimal system throughput at all the locations. The computer simulation results show that compared with the traditional single-beam, dual-beam and eight-beam schemes, the proposed scheme achieves 87.9%, 62.3%, and 50.6% improvement respectively in system throughput when the distance between the train and the BS is less than 125 m, achieves a similar system throughput of single beamforming when the distance is more than 125 m. The experimental results show that the proposed scheme has best system throughput whether the train is far away from or close to the BS, and is better adapted to high-speed railway environment.