一种基于机器学习的井间水驱优势通道识别方法  

A Method of Identifying Interwell Dominant Seepage Channel for Waterflooding Reservoirs Based on Machine Learning

作　　者：杨二龙[1,2] 陈柄君 董驰[1,2] 曾傲 张梓彤 YANG Erlong;CHEN Bingjun;DONG Chi;ZENG Ao;ZHANG Zitong(Northeast Petroleum University(MOE Key Laboratory of Enhanced Oil Recovery),Daqing,Heilongjiang 163318,China;SANYA Offshore Oil&Gas Research Institute,Northeast Petroleum University,Sanya,Hainan 572024,China)

机构地区：[1]东北石油大学·提高油气采收率教育部重点实验室 [2]东北石油大学三亚海洋油气研究院

出　　处：《钻采工艺》2025年第1期157-164,共8页Drilling & Production Technology

基　　金：海南省科技计划三亚崖州湾科技城联合项目“海上油气藏复杂流场分布规律研究”(编号:2021JJLH0059);黑龙江省博士后科研启动资金项目“大庆油田二类油层聚合物驱窜聚动态识别与优化控制”(编号:LBH-Q21012)。

摘　　要：井间优势渗流通道的形成受多方面的因素综合影响,识别过程中需要分析的因素众多、过程复杂,最直观可靠的做法是通过剖面测试数据结合生产动态分析来判定,或者通过措施见效井来验证是否存在优势渗流通道,但是实际生产中剖面测试数据量不足,措施见效井分析结果又属于后验知识,时效性差,导致识别的精度和效率较低。因此,本文以大庆油田特高含水典型区块M区块为例,结合主控因素分析方法构建特征参数集,应用粒子群算法(PSO)优化深度置信神经网络(DBN)的结构参数,通过逐层递推和全局优化融合、有监督和无监督学习算法融合提升模型性能,形成了一种基于机器学习算法的注采井间优势通道识别的方法。构建的优势通道识别PSO-DBN模型应用于典型区块,识别准确率比未经过优化的DBN神经网络模型预测准确率提高了2.8%,比MLP神经网络模型预测准确率提高了8.6%,通过增补无标注样本、实现有监督和无监督学习算法融合,可以进一步提升识别精度。The formation of interwell dominant seepage channels is influenced by a variety of factors,and the identification process requires analysis of numerous factors and is complex.The most intuitive and reliable approach is to determine the presence of such channels through profile testing data combined with production dynamic analysis,or to validate the existence of advantageous seepage channels through effective measures in test wells.However,in actual production,the amount of profile testing data is insufficient,and the analysis results from effective measures in test wells belong to retrospective knowledge,which has poor timeliness,leading to low precision and efficiency in identification.Therefore,this paper takes the typical ultra-high water cut block M of the Daqing Oilfield as an example,constructs a characteristic parameter set combined with the analysis of controlling factors,and applies the Particle Swarm Optimization(PSO)algorithm to optimize the structural parameters of the Deep Belief Network(DBN).By integrating layer-by-layer propagation and global optimization,and combining supervised and unsupervised learning algorithms,the model performance is enhanced,forming a method for identifying dominant channels in injection and production wells based on machine learning algorithms.The constructed PSO-DBN model for identifying dominant channels is applied to typical blocks,achieving an accuracy rate of identification that is 2.8%higher than that of the unoptimized DBN neural network model and 8.6%higher than that of the MLP neural network model.By supplementing unlabeled samples and implementing a fusion of supervised and unsupervised learning algorithms,the identification accuracy can be further improved.

关 键 词：特高含水油藏 井间优势通道 深度置信神经网络 算法融合 机器学习 

分 类 号：TE357.6[石油与天然气工程—油气田开发工程] TP181[自动化与计算机技术—控制理论与控制工程]