The optimization of random multiple contention access can significantly enhance the power of wireless gateways and is also a key prerequisite for edge computing applications. Aiming at the problem of low throughput of heterogeneous protocol multiple access systems in wireless IoT networks, an intelligent adaptive wireless multiple access method based on deep reinforcement learning is proposed. First, the access state is reinforced through channel perception, action feedback and loss minimization mechanism. Then, the improved proximal strategy optimization PPO algorithm is used to evaluate the optimal channel access strategy, and complementation with traditional TDMA and ALOHA protocols are achieved to reduce the collision of access time slots, thereby improving access resource utilization and network throughput. The results show that the improved algorithm can increase the throughput by 26.6% compared with the case without reinforcement learning, and by 2.6% compared with the DQN algorithm. It can effectively reduce the complexity of multiple access and significantly improve the multiple access performance of wireless gateways.