The hybrid networking of power line and wireless dual-mode communication can achieve complementary advantages, demonstrating significant application prospects in fields such as the Internet of Things for power systems and smart homes. To enhance the access flexibility and network resource utilization of dual-mode communication MAC algorithms, a superframe time slot resource allocation algorithm for dual-mode gateways based on deep reinforcement learning is proposed. Firstly, an interaction model between the dual-mode communication gateway and terminals is established, detailing the superframe structure and the specific execution steps of each phase. Secondly, key reward functions, state spaces, and action spaces for machine learning applications are defined, with the gateway monitoring and collecting parameters such as superframe access throughput. Through iterative learning and training, optimized parameters including the number and proportion of superframe time slots are obtained. Finally, simulations are conducted to verify the effectiveness and reliability of the proposed algorithm. Comparisons are made with fixed superframe structure and non-superframe structure algorithms, analyzing network throughput, average delay, packet loss rate under different algorithms, as well as the influence patterns of key parameters on system performance. Simulation results indicate that the proposed algorithm can effectively improve system performance in terms of throughput and delay, enabling flexible and efficient resource allocation.