Decision Transformer is one of the representative algorithms in the field of offline reinforcement learning for conditional sequence modeling. It can learn from past decisions and predict future action strategies, performing well in most tasks. However, recent research indicates its limitations in concatenating different trajectory segments, meaning it struggles to utilize suboptimal trajectories effectively. This paper combines Decision Transformer with knowledge graph link prediction, leveraging the strengths of both methods to enhance decision-making in complex environments. While Decision Transformer provides a robust framework for handling sequential data, knowledge graph link prediction offers structured representations of entity relationships and predicts potential links between entities, allowing the model to infer unobserved state transitions. By integrating knowledge graph predictions into the Decision Transformer model, we can enhance decision accuracy by leveraging additional contextual and relational information. This synergy enables more effective utilization of available data and better adaptation to diverse scenarios. Through experiments and evaluations, we have demonstrated the effectiveness of this integration approach in improving the performance of offline reinforcement learning models across various tasks and domains.