In order to solve the problem of resources allocation caused by operators sharing network resources, this paper proposes an online bandwidth resources allocation algorithm based on deep reinforcement learning(DRL). The algorithm maps multi-tenancy passive optical network(PON) system to DRL model. The DRL agent interacts with the environment to make decisions for each pending bandwidth request and the current remaining bandwidth, and constantly updates the policy parameters until the model converges, so as to complete the algorithm optimization. A simulation system is built to verify the feasibility of the algorithm. The simulation results show that the proposed algorithm can effectively improve the utilization of bandwidth resources.