Traditional Knowledge Graph Reasoning (KGR) methods often neglect the complex relations and structures within knowledge graphs, leading to low reasoning accuracy and making them inadequate for real-world AIOps scenarios. We propose a path-based KGR method called OpsKGR, which is based on reinforcement learning and introduce a variant of Graph Neural Network with a relation-aware neighborhood aggregation mechanism. This approach achieves neighborhood aggregation around the entity node, capturing additional path, relation, and structure information within the knowledge graph, thereby improving reasoning accuracy and robustness. We apply OpsKGR to a domain KG constructed using operation and maintenance data from the State Grid Corporation of China. Experimental results show that OpsKGR outperforms the state-of-the-art methods, with improvements of 2.6%, 2.8%, and 3.6% on MRR, Hits@1, and Hits@10 metrics. Extensive experiments on three additional public datasets further validate OpsKGR’s superiority over baselines, demonstrating its effectiveness and scalability in KGR.