基于BP神经网络的咸水黏度预测及其对渗流的影响  

Brine water viscosity prediction based on BP neural network and its effect on flow

作　　者：李涛[1,2] 美合日阿依·穆太力普[1,2] 薛福生 李延静 敬嘉珩 LI Tao;MEIHERIAYI Mutailipu;XUE Fusheng;LI Yanjing;JING Jiaheng(Engineering Research Center of Northwest Energy Carbon Neutrality,Ministry of Education,Urumqi,Xinjiang,830017,China;School of Electrical Engineering,Xinjiang University,Urumqi,Xinjiang,830017,China)

机构地区：[1]西北能源碳中和教育部工程研究中心,新疆乌鲁木齐830017 [2]新疆大学电气工程学院,新疆乌鲁木齐830017

出　　处：《油气地质与采收率》2025年第1期152-161,共10页Petroleum Geology and Recovery Efficiency

基　　金：新疆维吾尔自治区中央引导地方科技发展专项“新疆高碳排放行业碳中和关键技术及工程应用项目课题”(ZYYD2022C16);新疆维吾尔自治区重点研发专项课题“区域CCUS源汇匹配优化路径及二氧化碳捕集与强化驱油关键技术”(2022B01033-2);新疆维吾尔自治区自然科学基金青年项目“基于低温多效蒸馏技术(MED)的矿井水淡化中盐水热力学特性研究”(2021D01C089);国家自然科学基金项目“致密层CO_(2)地质封存强化采水多相界面演化与孔渗特性变化研究”(52366010);新疆维吾尔自治区高校基本科研项目“基于相平衡理论的CO_(2)与咸水溶解-酸化特性研究”(XJEDU2023P026)。

摘　　要：在碳中和背景下,采用CO_(2)咸水层封存技术来实现碳减排目标时,咸水黏度对储层中的CO_(2)-咸水两相渗流过程有着直接的影响。目前,基于压力影响的黏度预测方法仍有待完善。使用最小二乘法、BP神经网络和基于遗传算法的BP神经网络,将咸水黏度分别当作温度、质量摩尔浓度的二元函数以及温度、质量摩尔浓度、压力的三元函数优化了现有的计算方法,建立了考虑压力影响的黏度预测优化模型。在获得最佳的预测方式后,基于COMSOL软件的水平集方法系统分析了黏度对渗流的影响。研究结果表明,采用最小二乘法可以对现有的经验公式进行一定优化,但是效果不明显;采用二元BP神经网络可以将预测精度提高45.20%,考虑压力后采用三元BP神经网络可以将预测精度提高57.32%。因此,在实验数据充足的情况下,基于BP神经网络模型可以得到较大压力范围内可靠的咸水黏度值;由于经验公式法能够预测黏度变化趋势,在缺乏相应实验数据的情况下,可通过经验公式法获得咸水黏度值。此外,通过仿真结果可以发现,黏度会影响流体在流道的分布,进而影响流动速度,黏度比越大,出口平均速度波动越小且更快地趋于平稳;而且黏度比越大,残余水饱和度越小,越有利于驱替过程的进行,二者呈对数函数的关系。In the context of carbon neutrality,brine water viscosity directly influences the CO_(2)-brine water two-phase flow process in the reservoir when the technology of CO_(2) storage in brine water reservoirs is used to achieve carbon reduction targets.Currently,the viscosity prediction method based on the pressure effect still needs to be improved.In this study,the existing calculation methods were optimized by using the least squares method,BP neural network,and BP neural network based on a genetic algorithm.The brine water viscosity was taken as a binary function of temperature and molar concentration,as well as a ternary function of temperature,molar concentration,and pressure,respectively,and an optimized model of viscosity prediction considering the effect of pressure was established.Then,the optimal prediction was obtained,and the effect of viscosity on flow was systematically analyzed based on the level-set method of COMSOL software.The results show that the existing empirical formula could be optimized using the least squares method,but the effect is not apparent.After considering the pressure,the prediction accuracy can be improved by 45.2% using the binary BP neural network and by 57.32% using the binary BP neural network.Therefore,in the case of sufficient experimental data,reliable brine water viscosity values can be obtained in a wide range of pressures based on the BP neural network model.In the absence of the corresponding experimental data,the brine water viscosity values can be obtained by the empirical formula method because the empirical formula method can predict the trend of viscosity change.In addition,it can be found that the viscosity affects the distribution of the fluid in the flow channel through the simulation results,which in turn affects the flow velocity.A larger viscosity ratio indicates a minor fluctuation of the average velocity at the outlet and faster stability,and a larger viscosity ratio indicates smaller residual water saturation,which is more favorable for the replacement proc

关 键 词：咸水黏度 BP神经网络 压力 渗流模拟 CO_(2)咸水层封存 

分 类 号：TE312[石油与天然气工程—油气田开发工程]