To provide a fully automatic antenna adaptive adjustment scheme with advantages of better performance, wider coverage and lower maintenance cost, the key design technologies of intelligent adjustment system of adaptive antenna feed system of 5g-based station are studied from the perspective of signal radiation direction adjustment of antenna panel. An adaptive adjustment strategy for base-station antenna based on deep reinforcement learning is proposed. The adaptive antenna feed system designed with the proposed strategy can use telecom RSRP coverage map as a data source, and obtain the current state of the observed values to automatically analyze data and adjust the antenna panels. In a virtual environment, the system based on reinforcement learning is simulated and trained, and the results are in line with expectations.