Objective The development of primary logistics plans for refined oil currently relies heavily on logistics management software based on optimization models. In large-scale and complex logistics systems, issues such as supply-demand imbalances and insufficient transportation capacity frequently arise, often resulting in constraint conflicts within the models. As a result, operators must manually adjust numerous parameters, leading to high operational complexity and significant training costs, which hinder the efficient generation of feasible plans. Traditional approaches that depend on static optimization and manual parameter tuning struggle to adapt to an increasingly dynamic and complex business environment. Advancing parameter-tuning methods toward greater intelligence and adaptability is therefore essential for achieving efficient and flexible logistics planning.

Methods Based on a review of relevant domestic and international research, the status of primary logistics plan formulation methods for refined oil and the application of artificial intelligence in related fields was analyzed. Three main issues with current methods were identified: difficulty in generating feasible plans, low formulation efficiency, and unsatisfactory economic benefits. To address these, an intelligent parameter-tuning method for the optimization model of primary refined oil logistics was developed. This method comprised two core modules. First, a knowledge base for model parameter tuning was constructed using knowledge graph technology, in which structured knowledge quadruples were formed from model relaxation variables and parameter adjustment measures, ensuring interpretability, updatability, and inferability of the rules. Second, an intelligent parameter-tuning model was designed. Relaxation variables were introduced into the logistics optimization model to reconstruct the objective function and constraints, enabling the model to output “conflict positions” and “adjustment amplitudes”. A graph neural network algorithm was then integrated to classify problem types and infer parameter-tuning measures in complex logistics scenarios.

Results The intelligent parameter-tuning method, integrating knowledge graphs and graph neural networks, enhanced the optimization model’s capability to generate feasible plans under complex constraints, enabled automatic problem identification and intelligent parameter tuning, and improved the economic efficiency of the logistics plan.

Conclusion The proposed method offers an effective path for the intelligent upgrade of parameter-tuning techniques. Future development should be driven by multi-source data to support the transformation of refined oil logistics planning toward greater efficiency and intelligence.