The invention discloses an intelligent QoS routing optimization method and system based on deep reinforcement learning in an SDN environment, and the method comprises the following steps: expressing all streaming media services in a network as a service request set, and then for each request, searching a path meeting the network service quality from a streaming media server to a heterogeneous client; sequentially determining the route of each flow request, and finally constructing a multicast tree by adopting a QoS route optimization algorithm. For a network congestion link or a malicious node, the most suitable next node at present can be found for routing through a deep reinforcement learning method. By adopting the method of combining deep learning and reinforcement learning, the transmission delay of the video stream can be effectively reduced, and the accuracy of routing decision can be improved. Meanwhile, the design of a distributed control plane is adopted and can be realized in various network topologies, so that the network congestion can be avoided, the expandability of the network is improved, the interaction with a single controller is reduced, and the overall utilityof the network is improved.