In order to solve the problem of limited performance of traditional routing algorithms in software-defined networks (SDN) under large-scale and dynamically changing environments, this study proposed a routing optimization algorithm based on deep reinforcement learning. The algorithm realizes autonomous learning and dynamic adjustment of routing strategies by training intelligent agents, aiming to improve network bandwidth utilization efficiency and reduce transmission delay. It combines deep learning and reinforcement learning to process network status information and optimize routing decisions respectively to adapt to real-time changes in the network environment and ensure optimal performance under complex network conditions. Experimental results show that compared with traditional methods, this algorithm has achieved significant improvements in network throughput, latency and routing efficiency, effectively coped with the dynamics and uncertainty in SDN networks, and enhanced network performance and reliability. At the same time, the algorithm also realizes balanced allocation of network resources, further improving the overall allocation efficiency, which has important theoretical and practical value for building a smarter and more reliable network routing system.