耦合混油发展机理与数据修正的成品油管道混油浓度预测  

Prediction of mixed oil concentration in product oil pipeline coupling oil mixing mechanism and data correction

作　　者：杜渐 郑坚钦 夏玉恒 张秀玲[2] 徐宁 廖绮 涂仁福 梁永图[1] DU Jian;ZHENG Jianqin;XIA Yuheng;ZHANG Xiuling;XU Ning;LIAO Qi;TU Renfu;LIANG Yongtu(National Engineering Research Center for Pipeline Safety,China University of Petroleum(Beijing)//MOE Key Laboratory of Petroleum Engineering/Beijing Key Laboratory of Urban Oil and Gas Distribution Technology;PetroChina Planning&Engineering Institute;Engineering School,China University of Petroleum-Beijing at Karamay)

机构地区：[1]中国石油大学(北京)油气管道输送安全国家工程研究中心·石油工程教育部重点实验室·城市油气输配技术北京市重点实验室 [2]中国石油规划总院 [3]中国石油大学(北京)克拉玛依校区工学院

出　　处：《油气储运》2024年第7期796-808,共13页Oil & Gas Storage and Transportation

基　　金：国家自然科学基金资助项目“面向大规模成品油管网调度的数据解析与优化融合方法”,52202405。

摘　　要：【目的】顺序输送成品油管道不可避免会在相邻油品交界处产生混油段,影响站场调度计划与分输方案制定,准确预测混油段浓度分布是提高站场混油控制水平、降低混油处理能耗、防止油品质量事故的重要依据。然而,多维混油数值模型计算时间过长、难以应用于长输管道,传统数据驱动方法忽略混油发展机理过程,结果可能违背物理原理,准确度及可解释性差。【方法】通过分析混油发展机理,深入挖掘混油发展遵循的基本控制方程与初始边界条件,通过自动微分方式与深度学习模型耦合以构造物理指导的耦合损失函数,将模型预测结果约束至混油发展所对应的物理解空间内;再利用短管道初始数值解,基于管道递推滚动预测与数据修正方法,预测长输管道混油浓度分布。【结果】数值案例表明,所建模型相较传统数据驱动方法准确度更高,MAPE降低了91%,对数据依赖程度更低,RMSE波动程度随数据量变化降低了60%;相较Fluent数值仿真方法,计算成本降低至原先的12%;现场工程化应用实例表明,相较Taylor模型与改进Taylor模型,所提出的工程化应用框架的MAE分别降低了71%、58%,可实现长输管道混油浓度分布高效求解。【结论】耦合混油发展机理与数据修正的成品油管道混油浓度预测方法可准确快速预测长输管道混油浓度分布,指导站场混油接收方案制定,有效提高混油控制智能化水平。[Objective]Batch pipelining of product oils inevitably leads to oil mixing at the junction of oils transmitted in sequential batches within the pipelines.This disruption affects the formulation of station dispatching plans and oftake schemes.Therefore,accurately predicting the concentration distribution of oil mixing sections is considered a crucial foundation for enhancing the control of oil mixing,reducing energy consumption for mixed oil treatment,and preventing oil quality incidents at stations.Despite this,multi-dimensional numerical models of oil mixing have exhibited several drawbacks.These include lengthy calculation processes,ineficiencies in long-distance pipeline applications,neglect of the oil mixing progression mechanism process due to traditional data-driven approaches,and violations of physical principles,low accuracy,and poor interpretability in their results.[Methods]Through analyzing the oil mixing progression mechanism,this study delved into the fundamental control equations and initial boundary conditions relevant to oil mixing progression.A physics-guided coupling loss function was constructed by coupling the automatic differential method with a deep learning model,enabling the confinement of model prediction results within the corresponding physical solution space of oil mixing progression.Following this,the initial numerical solutions for short pipeline sections were used to predict the mixed oil concentration distribution throughout long-distance transmission pipelines based on the recursive rolling prediction and data correction methods.[Results]The numerical examples demonstrated a higher accuracy of the established model than traditional data-driven methods,manifesting a notable 91%decrease in MAPE.This model also displayed reduced dependency on data and alleviated Root Mean Square Error(RMSE)fluctuations by 60%with changes in data size.Moreover,the computational expenses were minimized to a mere 12%of those incurred by the Fluent numerical simulation method.In practical engineering app

关 键 词：成品油管道 混油浓度预测 物理信息神经网络 基本控制方程 滚动预测 

分 类 号：TE832[石油与天然气工程—油气储运工程]