Abstract: Objective The formulation of refined oil primary logistics plans heavily relies on logistics management systems driven by optimization models. However, in large-scale and complex logistics networks, frequent issues such as supply–demand imbalances and insufficient transport capacity often result in constraint conflicts. Consequently, extensive manual parameter adjustments are required, leading to high operational thresholds and training costs, and making it difficult to efficiently generate feasible plans. Traditional approaches based on static optimization and manual adjustment have become inadequate in increasingly dynamic and complex environments. Therefore, advancing parameter adjustment methods toward intelligent and adaptive solutions is essential for achieving efficient and flexible logistics planning. Methods This study reviews relevant research both domestically and internationally, analyzing the current state of logistics planning methods and the application of artificial intelligence in related domains. It proposes an intelligent parameter adjustment framework for refined oil primary logistics optimization model, aimed at addressing three key challenges: difficulty in generating feasible plans, low planning efficiency, and suboptimal economic performance. The framework includes two core modules: (1) construction of a model adjustment knowledge base using knowledge graph techniques, encoding relaxation variables and adjustment strategies into structured, interpretable, and inferable knowledge quadruples; and (2) development of an intelligent parameter adjustment model that introduces relaxation variables into the optimization formulation, enabling identification of conflict points and adjustment magnitudes. A graph neural network is applied to support classification and reasoning of adjustment strategies in complex logistics scenarios. Results By integrating knowledge graphs with graph neural networks, the proposed method enhances the model’s ability to generate feasible plans under complex constraints. It enables automatic identification of model problem and intelligent parameter adjustments, improving the economic performance of logistics operations. Conclusion The proposed approach offers a promising path toward intelligent and adaptive parameter adjustment, supporting the transition of refined oil logistics planning toward greater efficiency and smarter decision-making under multi-source data environments.