基于机器学习的油气水三相混输管道起点压力预测  

Prediction of Initial Pressure for Oil-Gas-Water Three-Phase Mixed Transportation Pipeline Based on Machine Learning

作　　者：牛鹏涛 杜渐 潘诗元 徐宁 耿宇 李焯超 梁永图[1] 李欣泽 NIU Pengtao;DU Jian;PAN Shiyuan;XU Ning;GENG Yu;LI Zhuochao;LIANG Yongtu;LI Xinze(National Engineering Laboratory for Pipeline Safety·MOE Key Laboratory of Petroleum Engineering·Beijing Key Laboratory of Urban Oil and Gas Distribution Technology·China University of Petroleum(Beijing);China University of Petroleum(Beijing)Karamay Campus)

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实验室·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]中国石油大学(北京)克拉玛依校区

出　　处：《油气田地面工程》2025年第2期36-41,48,共7页Oil-Gas Field Surface Engineering

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);新疆维吾尔自治区自然科学基金面上项目“二氧化碳管道停输再启动温压协同变化机理与安全控制理论研究”(2023D01A19)。

摘　　要：准确的油气水三相混输管道压力计算是油气集输管道优化设计和安全管理的关键,目前缺乏完全普适的油气水三相混输管道的压力预测方法。为了提高现场管理水平,针对西北S油田油气水三相混输管道建立了基于机器学习算法的起点压力预测模型。针对现场某混输管道建立PIPESIM模型,基于实际的管内流体物性、管道边界条件等参数,模拟了大量运行数据;为防止模型过拟合、提高模型的泛化能力,将数据随机打乱,并基于预测误差损失函数训练模型。以RMSE、MAE、R2为评价指标,对比了SVM、DT、RF、XGBoost、BP和GA-BP这6种回归预测模型。研究结果表明:GA-BP神经网络取得了该混输管道起点压力最准确的预测结果,其决定系数达到了0.9996,RMSE和MAE分别为1.5319和1.1420,表明模型的拟合能力优良,具有优越的学习能力和自适应拟合能力。该预测模型可用于油田油气水三相混输管道的压力监测,有效提高油田智能化运行管理水平。Accurate calculation of pressure in oil-gas-water three-phase mixed transportation pipelines is crucial for the optimization design and safety management of oil and gas gathering pipelines.Currently,there is a lack of universally applicable pressure prediction methods for such three-phase mixed transportation pipelines.In order to improve on-site management capabilities,a machine learning-based initial pressure prediction model for an oil-gas-water three-phase mixed transportation pipeline is established in S Oilfield in northwest China.A PIPESIM model is developed for a specific mixed transportation pipeline on-site,a substantial amount of operational data is simulated based on actual parameters such as fluid properties and pipeline boundary conditions.To prevent overfitting of the model and improve its generalization ability,the data is randomly shuffled,and the model is trained based on a predictive error loss function.Using RMSE,MAE,and R2 as evaluation metrics,six regression prediction models,namely SVM,DT,RF,XGBoost,BP,and GA-BP are compared.The results indicate that the GA-BP neural network achieves the most accurate prediction of the initial pressure in the mixed transportation pipeline,with a determination coefficient reaching 0.9996.The RMSE and MAE are 1.5319 and 1.1420,respectively,demonstrating excellent fitting capabilities,as well as the superior learning and adaptive fitting abilities of the model.This prediction model can be applied to the pressure monitoring of oil-gas-water three-phase mixed transportation pipelines in oilfields,effectively enhancing the level of intelligent operation management in the oilfield.

关 键 词：混输管道 油气水三相流 机器学习 预测模型 压力预测 GA-BP 

分 类 号：TE973[石油与天然气工程—石油机械设备] TP181[自动化与计算机技术—控制理论与控制工程]