A Combinatorial Machine Learning Models -Based Method for Predicting the Viscosity-Temperature Relationship of Crude Oil

To address the complexity of predicting the viscosity-temperature relationship in the pipeline transportation of highly viscous and gel-prone crude oil, this paper innovatively proposes a combinatorial Machine Learning Models that integrates the strengths of multiple algorithms. By leveraging the DBSCAN and the XGBoost, the robustness and generalization capability of viscosity-temperature characteristic prediction are significantly enhanced. The results demonstrate that this model achieves mean absolute percentage errors (MAPE) of 5.39% and 1.77% for the prediction of consistency coefficient K and flow behavior index n, respectively, while keeping the viscosity prediction error within 10%. The model also exhibits outstanding performance in sensitivity to data volume and cross-condition adaptability validation, providing a high-precision and transparent prediction tool for crude oil flow assurance.