In order to enhance the competitiveness of business and customer experience, the number of performance sensitive types of business such as low latency and high reliability is rapidly increasing. Performance sensitive services pose significant challenges to SDN routing performance. To meet the network performance requirements of performance sensitive business, this paper designs a multi-dimensional collaborative SDN network routing optimization model. Secondly, a routing optimization mechanism for performance sensitive services in power communication networks is proposed. This mechanism mainly includes four steps: performance sensitive business classification, generating corresponding routing strategies based on business types, real-time collection of QoS information, and generating routing optimization strategies based on reinforcement learning algorithms. In the performance analysis section, the feasibility and implement ability of the routing optimization strategy in this paper were verified to have good performance and application value.