摘要
针对天然气采气管线冬季冻堵事故频发这一问题,以井口压力、井口温度、日产气量、日产水量和天然气相对密度为输入,以甲醇理论加注量为输出,利用BP(back propagation)人工神经网络技术对采气管线甲醇理论加注量进行了预测,采用Sobol灵敏度分析找出了显著影响甲醇理论加注量的关键参数,并对注醇分配管网进行了优化。结果表明:5-16-12-1型BP人工神经网络经过2764次迭代后,它的训练样本、验证样本、测试样本均方误差分别为0.0055、0.0072和0.0085,均小于容许收敛误差限0.0100,而其决定系数亦高达0.9995、0.9986、0.9964,表现出良好的相关性;井口温度、天然气相对密度和井口压力对甲醇理论加注量影响较大,而井口压力、井口温度、日产气量、日产水量与其他参数之间可能存在明显交互作用;靖99-66、靖99-65、靖99-66H2和靖98-65在放射状和环状管网模式下甲醇实际加注量小于甲醇理论加注量,存在欠注现象,而新北5站所有气井在树枝状管网模式下甲醇实际加注量都大于甲醇理论加注量,起到了预防冻堵的目的。研究结果可为天然气采气管线注醇过程的节能降耗、提质增效提供必要的理论支撑和数据来源。
Aiming at the problem of frequent occurrence of freezing blocking accidents in natural gas production pipelines during winter,the technology of BP(back propagation)artificial neural networks was utilized to predict the theoretical quantity of methanol injection as outputs,considering the wellhead pressure,the wellhead temperature,the daily gas production,the daily water production,and the relative density of natural gas as inputs.The Sobol sensitivity analysis was used to find the key parameters that might significantly affect the theoretical quantity of methanol injection.Meanwhile,the distribution network relevant to methanol injection was also optimized.The prediction model of the 5-16-12-1 BP artificial neural networks evolves 2764 epochs until the mean square errors of training samples,verification samples,and test samples are 0.0055,0.0072,and 0.0085 respectively,which are less than the allowable convergence error limit of 0.0100.A good correlation is presented as the determination coefficients reaching 0.9995,0.9986,and 0.9964 respectively.The wellhead temperature,the relative density of natural gas,and the wellhead pressure have a greater impact on the theoretical quantity of methanol injection,while there may be significant interactions between the wellhead pressure,wellhead temperature,gas production per day,water production per day and other factors.On the conditions of the radial and circular pipe network mode,there is a shortage as the practical quantity of methanol injection is less than the theoretical quantity of methanol injection for Jing 99-66,Jing 99-65,Jing 99-66H2,and Jing 98-65.However,the practical quantity of methanol injection is more than the practical quantity of methanol injection for all gas wells of Xinbei Station 5 under the dendritic pipe network mode to achieve the goal of preventing freezing blockage.The research results could provide an essential theoretical basis and data source for the consumption reduction and benefits increment of the methanol injection process of natural gas production pipelines.
作者
梁生荣
李文君
翁军利
郝丽
杨琴
夏勇
刘钊
范峥
Liang Shengrong;Li Wenjun;Weng Junli;Hao Li;Yang Qin;Xia Yong;Liu Zhao;Fan Zheng(College of Chemistry&Chemical Engineering,Xi’an Shiyou University,Xi’an,Shaanxi,China;The First Gas Plant,PetroChina Changqing Oilfield Company,Yulin,Shaanxi,China)
出处
《石油与天然气化工》
CAS
CSCD
北大核心
2020年第6期45-52,共8页
Chemical engineering of oil & gas
基金
中国国家留学基金“高能效化工过程中的气液污染控制”(201908610135)
西安石油大学研究生创新与实践能力培养项目“上古气田注醇分配方法研究”(YCS19113080)。
关键词
冻堵
注醇
BP人工神经网络
预测
Sobol灵敏度分析
分配管网
优化
freezing blockage
methanol injection
BP artificial neural networks
prediction
Sobol sensitivity analysis
distribution network
optimization
