Q-Learning-Based Dynamic Traffic Routing Optimization for Adaptive Path Selection in Complex Networks

With the continuous expansion of network scale and the increasing dynamics of traffic patterns, traditional static or rule-based routing strategies are showing performance limitations in handling complex link states and bursty traffic. To address the challenges of real-time path selection and traffic scheduling in dynamic network environments, this paper proposes a Q-Learning-based dynamic routing optimization method. The proposed approach constructs a state representation mechanism incorporating link load, path delay, and topological structure, and introduces a multi-objective reward function to guide the policy learning process, enabling the agent to adaptively select routing paths under varying conditions. A series of experiments were conducted on the Mininet platform, comparing the proposed method with classical strategies such as Dijkstra, ECMP, and Random routing. Experimental results demonstrate that the proposed method achieves significant improvements in average delay, link utilization, and congestion control. It shows strong robustness and scalability, especially in medium- to large-scale networks and under bursty traffic scenarios. The study further indicates that the Q-Learning model can achieve efficient policy learning and optimization without relying on global network knowledge. This work provides a practical solution for intelligent routing and lays a theoretical and experimental foundation for the development of adaptive, high-performance network control systems.