Due to the theoretically unconditional security offered by quantum communication, it has broad application prospects in new power systems. However, the integration of quantum communication into these systems faces the challenge of reconciling the low quantum key generation rate with the high data transmission demands of power operations. This issue is particularly exacerbated by transmission distance limitations, necessitating the support of relay technology in the construction of quantum communication networks within power systems, which further amplifies the problem of low quantum key generation rates. To address these challenges, this paper models a quantum communication network architecture for power systems based on trusted relay nodes and designs a reinforcement learning-based quantum key distribution (QKD) routing algorithm. This algorithm aims to reduce network congestion and minimize routing hops by selecting the next node for the quantum key distribution task using reinforcement learning. Experimental results demonstrate that this algorithm can effectively improve the quantum key generation rate.