中国石油大学学报（自然科学版）2025，Vol.49Issue(2)：223-230,8.DOI:10.3969/j.issn.1673-5005.2025.02.022
耦合物理约束和数据模型改进的成品油管道混油长度预测
Coupling physical constraints with enhanced data-driven model for predicting contamination product length of multi-product pipeline
摘要
Abstract
When transporting oil products in multi-product pipeline sequentially,the contamination segment will inevitably form between adjacent batches.Predicting contamination product length accurately is of great importance for oil quality control and reducing the cost of contamination product handling.However,current works lack comprehensive consideration of influen-cing factors and full excavation of transmix migration mechanism,leading to dependence on amount of available data and poor robustness and accuracy.In this work,the transmix migration process is analyzed to extract feature variables,aiming to consid-er the influencing mechanism of factors on contamination product length.The customized network is designed for model fusion to improve interpretability and nonlinear correlation ability.The hierarchical training strategy based on coupling loss function is proposed to optimized model parameters,aiming to improve convergence effect and robustness.Results from historical contami-nation product small batch data suggest that,the proposed model achieves better accuracy and efficiency than conventional ma-chine learning models,with a 64%reduction of RMSE.Compared to current work,the distribution interval and average value of error predicted by the proposed are reduced by 46%and 21%,indicating a significant improvement of robustness and accura-cy.The proposed model is capable to predict the contamination product length of multi-product pipelines with different operat-ing characteristics accurately,thus improving the intelligence of contamination product control technology.关键词
成品油管道/混油长度预测/物理约束/梯度优化机制/深度学习模型Key words
multi-product pipeline/contamination product length predicting/physical constraint/gradient enhanced mecha-nism/deep learning model分类
能源科技引用本文复制引用
杜渐,郑坚钦,蔡庆文,徐宁,夏玉恒,涂仁福,梁永图..耦合物理约束和数据模型改进的成品油管道混油长度预测[J].中国石油大学学报（自然科学版）,2025,49(2):223-230,8.基金项目
国家自然科学基金项目(52202405) （52202405）
