物理信息神经网络驱动的井筒温度场求解方法  

A method for solving wellbore temperature field driven by physical information neural network

作　　者：刘雪琪 王志远[1] 魏强 王敏 孙小辉 王雪瑞[4] 张剑波 尹邦堂[1] 孙宝江[1] Liu Xueqi;Wang Zhiyuan;Wei Qiang;Wang Min;Sun Xiaohui;Wang Xuerui;Zhang Jianbo;Yin Bangtang;Sun Baojiang(School of Petroleum Engineering,China University of Petroleum,Shandong Qingdao 266580,China;Lunnan Oil and Gas Production Management Area,PetroChina Tarim Oilfield Company,Xinjiang Korla 841000,China;Kela Oil and Gas Production Management Area,PetroChina Tarim Oilfield Company,Xinjiang Korla 841000,China;College of Computer Science and Technology,China University of Petroleum,Shandong Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)石油工程学院,山东青岛266580 [2]中国石油塔里木油田公司轮南采油气管理区,新疆库尔勒841000 [3]中国石油塔里木油田公司克拉采油气管理区,新疆库尔勒841000 [4]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《石油学报》2025年第2期413-425,共13页Acta Petrolei Sinica

基　　金：国家自然科学基金重点项目“超深气井生产管柱泄漏精准识别与压力管控”(No.52434002);国家自然科学基金基础科学中心项目“超深特深层油气钻采流动调控”(No.52288101);国家自然科学基金联合基金项目“深水控压钻完井地层-井筒多场耦合机理与压力调控”(No.U21B2069);国家重点研发计划项目“压井处置智能决策支持与控制系统”(2023YFC3009204);山东省重大科技创新工程项目“深水复杂钻井多相流动模拟关键技术与监测装备”(2022CXGC020407)资助。

摘　　要：深水、深层油气钻探过程中对井筒温度场计算的实时性要求高,高精度、高效率的井筒温度场求解方法是精确计算流体物性、精细保障井筒流动安全的关键。将井筒温度场模型以损失函数形式嵌入神经网络,利用自适应权重和自适应学习率的优化方法提高训练效率,建立了物理信息神经网络驱动的井筒温度场求解方法,分析了钻井和气井测试期间井筒温度的瞬态变化。研究结果表明:钻井期间,与有限差分算法相比,钻杆温度和环空温度的平均误差分别为0.847%和0.725%,井底温度和井口温度的平均误差分别为0.162%和1.047%,计算效率提高约150倍;与现场实测数据对比,物理信息神经网络驱动的预测解与有限差分数值解的平均误差分别为2.16%和2.27%,规避偏微分方程的截断误差有助于提高模型精度;气井测试2 d内,天然气水合物生成风险的推断时间为0.7281 s,该方法可应用于水合物生成区域的快速预测。提出的求解方法在保证计算精度的同时,可大幅度提高计算速度。In the drilling process of deepwater and deep oil and gas,there is a high demand for real-time calculation of the wellbore temperature field.Therefore,a high-precision and high-efficiency wellbore temperature field solution method is the key to accurately calculate fluid properties and precisely guarantee the safety of wellbore flow.In this study,a wellbore temperature field model is embedded into the neural network in the form of loss function,and the optimization method of self-adaptive weight and self-adaptive learning rate is used to improve the training efficiency.Further,the paper establishes a method for solving the wellbore temperature field driven by physical information neural network,and analyzes the transient changes in wellbore temperature during drilling and gas well testing.The results show that during drilling,the average errors of drill pipe temperature and annular temperature are 0.847%and 0.725%,respectively,and those of bottom hole temperature and wellhead temperature are 0.162%and 1.047%,respectively,from which it can be seen that the computational efficiency is improved by about 150 times when compared with the finite difference algorithm.Compared with the field measurments,the average errors of the predicted solution driven by the physical information neutral network and the finite difference numerical solution are 2.16%and 2.27%,respectively,and the model accuracy can be improved by avoiding the truncation errors in partial differential equations.During the gas well testing for two days,the inferred time for the risk of natural gas hydrate formation is 0.7281 s,and this method can be applied to quickly predict the hydrate formation areas.In conclusion,the proposed solution method can not only ensure the calculation accuracy,but also significantly improve the computational speed.

关 键 词：钻井 气井测试 井筒温度 物理信息神经网络 快速预测 

分 类 号：TE21[石油与天然气工程—油气井工程]