融合自适应变异策略与差分进化算法的油藏自动历史拟合方法  

Automatic reservoir history matching method based on adaptive mutation strategy and differential evolution algorithm

作　　者：张金鼎 张凯[1,2] 张黎明[1] 刘丕养[2] 陈旭 ZHANG Jinding;ZHANG Kai;ZHANG Liming;LIU Piyang;CHEN Xu(School of Petroleum Engineering,China University of Petroleum(East China),Qingdao City,Shandong Province,266580,China;School of Civil Engineering,Qingdao University of Technology,Qingdao City,Shandong Province,266033,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]青岛理工大学土木工程学院,山东青岛266033

出　　处：《油气地质与采收率》2025年第2期152-162,共11页Petroleum Geology and Recovery Efficiency

基　　金：国家自然科学基金项目“油藏开发智能实时优化”(52325402);“基于强化学习的离线-在线交互式油藏开发生产实时优化方法”(52274057);“基于迁移学习的油藏开发注采优化方法研究”(52074340);“基于电磁支撑剂的水力压裂裂缝监测理论与方法”(51874335);国家重点研发计划“CO_(2)驱油及封存安全监测技术”(2023YFB4104200);中国海油重大科技项目“基于代理模型的海上油藏智能注采流场实时调控优化方法研究”(CCL2022RCPS0397RSN)。

摘　　要：差分进化算法作为一种经典的进化算法,具有全局搜索能力、便于实现、无需梯度等优势,在油藏自动历史拟合中广泛应用,但算法中参数的设置对历史拟合结果影响较大,在高维问题中存在着收敛停滞的问题。为解决上述难题,提出一种融合自适应变异策略与差分进化算法的油藏自动历史拟合方法。首先,基于主成分分析方法对油藏模型的高维参数进行降维,将降维后的参数作为差分进化算法中调整的参数,以压缩变量的搜索空间,提升算法搜索效率;其次,结合自适应变异策略与差分进化算法,借助于算法搜索过程中的历史经验指导当前种群的更新,当种群个体停止收敛时,则切换差分进化算法的变异策略,改变种群的迭代更新方式,以此避免油藏参数停止优化调整的情况;此外,为使更新后模型参数与先验分布特征保持一致,应用分位数变换策略转换更新后参数的分布情况,将非高斯分布的数据变换为高斯分布,使更新后的模型更加符合实际地质参数的约束条件。提出算法在三维油藏模型上进行测试验证,结果表明:相比传统的差分进化算法框架,改进的差分进化算法不仅能够提升历史拟合求解的收敛效果,而且反演的油藏模型参数更加符合实际地质特征,在相同的计算条件下,可获得更优的历史拟合模型,数据拟合效果更显著。As a classical evolution algorithm,the differential evolution algorithm has the advantages of global search ability,easy implementation,and no gradient.It has been widely used in automatic reservoir history matching.However,the setting of parameters in the algorithm has a significant influence on the result of history matching,and there is convergence stagnation in high-dimensional problems.In order to solve the above problems,an automatic reservoir history matching algorithm was proposed based on the adaptive mutation strategy and differential evolution algorithm.Firstly,based on the principal component analysis method,the high-dimensional parameters of the reservoir model were reduced,and the reduced parameters were used as the parameters adjusted in the differential evolution algorithm to compress the search space of the variables and improve the search efficiency of the algorithm.Secondly,based on the adaptive mutation strategy and differential evolution algorithm,the historical experience in the search process of the algorithm was used to guide the update of the current population.When the individual of the population stopped converging,the mutation strategy of the differential evolution algorithm was switched to change the iterative update mode of the population to avoid the situation that the reservoir parameters stopped optimization and adjustment.In addition,to make the updated model parameters consistent with the prior distribution characteristics,the quantile transformation strategy was applied to transform the distribution of the updated parameters,and the data of non-Gaussian distribution was transformed into Gaussian distribution so that the updated model was more in line with the constraints of the actual geological parameters.The proposed algorithm was tested and verified on a three-dimensional reservoir model.The results show that compared with the traditional differential evolution algorithm framework,the improved differential evolution algorithm can improve the convergence effect of the history

关 键 词：油藏数值模拟 自动历史拟合 差分进化算法 自适应方法 分位数变换 

分 类 号：TE319[石油与天然气工程—油气田开发工程]