Abstract: To solve the problem of multi-user spectrum resource allocation under hybrid overlay/underlay spectrum sharing mode, a dynamic spectrum allocation model under hybrid spectrum sharing is constructed, and a multi-user dynamic spectrum allocation algorithm based on Q learning is proposed. Aiming at maximizing the total throughput of secondary users without causing harmful interference to primary users, the proposed algorithm establishes some virtual secondary users (SUs) corresponding to the actual SUs, and treats them as agents. Through interactive learning with the environment, channels and sharing patterns allocated to SUs are selected preliminarily. The channel allocation strategy is adjusted by spectrum allocation system according to the conflict states and the learn results of each agent until there is no conflict between SUs. Simulation results show that according to the channel states of the previous time slot and the transmission rate requirements of the SUs without any channel detection and prior knowledge of the channels, the dynamic channel allocation and spectrum sharing patterns can be determined. The proposed algorithm can also effectively avoid conflicts between SUs and reduce conflicts between primary users and SUs so that it effectively improves the total throughput of SUs.