SDN provides network managers with a global network view and programmability, making it more convenient for network managers to carry out security management. In recent years, machine learning has innovated many fields of computer science. The reinforcement learning method has emerged as an effective solution to sequential decision-making and control problems. Its method is widely used in the field of automation control. The computer network is becoming more and more complex and changeable. In the face of the emerging technology of IoT and the threat of attack and intrusion, the demand for adaptive security control is increasingly urgent. The reinforcement learning method is a feasible solution to achieve adaptive control. Based on the SDN framework, this paper develops and uses reinforcement learning applications for SDN network security management. For the existing problems, we improve the structure of D3QN and use the improved D3QN deep reinforcement learning agent to learn and mitigate apt attacks. Finally, the experimental results are evaluated, and the convergence results of the improved algorithm model are given, which shows the availability of reinforcement learning methods for adaptive threat mitigation in the SDN environment.