Abstract: Aiming at the problems of slow routing convergence, poor service adaptability and low link utilization of power dispatching data networks under the new power system, this paper proposes an algorithm system including three core modules: service priority-aware OSPF route selection and incremental convergence, multi-link collaborative Dueling DQN dynamic traffic scheduling, and adaptive weight-based MPLS VPN dynamic bandwidth allocation. The results of OPNET simulation and 6-month field verification on a prefecture-level power dispatching data network show that, the proposed algorithm system can shorten the routing convergence time by 90.7%, reduce the end-to-end delay of core services by 76.2%, improve the link bandwidth utilization by 128.6%, and achieve a 100% QoS guarantee rate for core services. This scheme provides a low implementation risk and cost-effective solution for the performance improvement of power dispatching data networks.