基于GA-BP理论的储层视裂缝密度地震非线性反演方法
|
摘要
基于GA-BP理论,将自适应遗传算法与人工神经网络技术(BP算法)有机地相结合,形成了一种储层裂缝自适应遗传-神经网络反演方法.这种新的方法是由编码、适应度函数、遗传操作及混合智能学习等组成,即在成像测井裂缝密度数据约束下,通过对目标问题进行编码(称染色体),然后对染色体进行选择、交叉和变异等遗传操作,使染色体不断进化,从而快速获得全局最优解.在反演执行过程中,利用地震数据和成像测井裂缝密度数据之间的非线性映射关系建立训练样本,将GA算法与BP算法有机地结合,优化三层前向网络参数;或将GA与ANFIS相结合,优化ANFIS网络参数.并采用GA算法与TS算法(Tabu Search)相结合的自适应混合学习算法,该学习算法自始至终将GA和BP两种算法按一定的概率比例进行,其概率自适应变化,以达到混合算法的均衡.这种混合算法提高了网络的收敛速度和精度.我们分别利用两个研究地区的6井和1井成像测井裂缝密度数据与地震数据之间的非线性映射关系建立训练样本,对过这两口井的测线的地震数据进行反演,获得了视裂缝密度剖面,视裂缝密度剖面上裂缝分布特征符合沉积相分布特征和岩石力学性质的变化特征.这种视裂缝密度剖面含有储层裂缝的定量信息,其误差可为油气勘探开发实际要求所允许.因此,这种新的方法优于只能作裂缝定性分析的常规裂缝地震预测方法,具有广阔的应用前景.
Based on GA-BP theory,by self-adaptation genetic algorithms united with artificial neural network technique(BP algorithms) organically,one kind of self-adaptation genetic - neural network technique for reservoir fracture inversion has been formed.This new technique is combined with the coding,adaptability function,genetic operation,hybrid intelligent training and so on.Under constraint of the fracture density measurements from imaging logging,by means of encode to the object problem(chromosome),then the genetic operations such as selective,cross and variation are chose to chromosome,causing the chromosome continuously evolution,thereby a global optimal solution is found speedily.During the inversion,the training specimen is established by utilizing nonlinear mapping relationship between seismic data and fracture density measurements.Organically combine GA algorithm with BP algorithm to optimize the three-level forward network parameter,either another combine GA with ANFIS to optimize ANFIS network parameter.Moreover self-adaptation hybrid training is adopted by combined GA algorithm with TS algorithm(Tabu search).The hybrid training algorithms take up with two algorithms of GA and BP according to the specified proba bility proportion all the way.Such the balance of the hybrid algorithm has attained because the probability varied self-adaptation.The hybrid algorithm has high accuracy and speed of convergence there by its effect of inversion is obviously.We utilize the nonlinear mapping relationship between seismic data and fracture density measurements from imaging logging of 6# wells and 1# well in two research areas to establish the training specimen respectively.The inversion has been processed to the seismic data of survey passing this two wells obtaining apparent fracture density section.The distribution feature of fracture on the apparent fracture density section corresponds to the distribution feature of sedimentary facies and the variation feature of rock mechanics nature.The apparent fracture density section has to keep in quantitative measurements of reservoir fracture and its error may be permitted upon the practice requirement of exploration and production for oil and gas.Hence,this new technique of seismic nonlinear inversion is superior to the conventional technique only to act as the qualitative analysis for feature prediction from seismic data.It has wide application foreground.
Based on GA-BP theory,by self-adaptation genetic algorithms united with artificial neural network technique(BP algorithms) organically,one kind of self-adaptation genetic - neural network technique for reservoir fracture inversion has been formed.This new technique is combined with the coding,adaptability function,genetic operation,hybrid intelligent training and so on.Under constraint of the fracture density measurements from imaging logging,by means of encode to the object problem(chromosome),then the genetic operations such as selective,cross and variation are chose to chromosome,causing the chromosome continuously evolution,thereby a global optimal solution is found speedily.During the inversion,the training specimen is established by utilizing nonlinear mapping relationship between seismic data and fracture density measurements.Organically combine GA algorithm with BP algorithm to optimize the three-level forward network parameter,either another combine GA with ANFIS to optimize ANFIS network parameter.Moreover self-adaptation hybrid training is adopted by combined GA algorithm with TS algorithm(Tabu search).The hybrid training algorithms take up with two algorithms of GA and BP according to the specified proba bility proportion all the way.Such the balance of the hybrid algorithm has attained because the probability varied self-adaptation.The hybrid algorithm has high accuracy and speed of convergence there by its effect of inversion is obviously.We utilize the nonlinear mapping relationship between seismic data and fracture density measurements from imaging logging of 6# wells and 1# well in two research areas to establish the training specimen respectively.The inversion has been processed to the seismic data of survey passing this two wells obtaining apparent fracture density section.The distribution feature of fracture on the apparent fracture density section corresponds to the distribution feature of sedimentary facies and the variation feature of rock mechanics nature.The apparent fracture density section has to keep in quantitative measurements of reservoir fracture and its error may be permitted upon the practice requirement of exploration and production for oil and gas.Hence,this new technique of seismic nonlinear inversion is superior to the conventional technique only to act as the qualitative analysis for feature prediction from seismic data.It has wide application foreground.
引文
[1]李勇,李正文.高分辨率非线性反演方法及应用研究.天然气工业,2004,24(3):58~60.Li Y,Li Z W.High resolution nonlinear inversion method andits application.Natural Gas Industry,2004,24(3):58~60.
[2]王小平,曹立明.遗传算法———理论、应用与软件实现[M].西安:西安交通大学出版社,2002.Wang X P,Cao L M.Genetic algorithms—theory,application,software implement[M].Xi an:Xian Jiaotong UniversityPress,2002.
[3]刘勇,康立山,等.非数值并行计算(第2册)———遗传算法[M].北京:科学出版社,1995.Liu Y,Kang L S,et al.Non-numerical value parallel compu-ting(Ⅱ)———Genetic algorithms[M].Beijing:Science press,1995.
[4]焦李成.神经网络的应用和实现[M].西安:西安电子科技大学出版社,1993.Jiao L C,Neural network application and implement[M].Xian:XIDIAN university press,1993.
[5]A.И.加卢什金著.神经网络理论[M].阎平凡译.北京:清华大学出版社,2003.AlexanderИG.Neural network theory.Yan P F translating[M].Beijing:Tsinghua university press,2003.
[6]Jang J S R.ANFIS:Adaptive-Network-Based Fuzzy InferenceSystem[J].IEEE Transaction on System,Man and Cybernet-ics,1993,23(3):665~685.
[7]Goldberg D E.Genetic Algorithms in Search,Optimization,andmachine Learning[J].Reading,MA,Addison-Wesley,1989.
[8]Whitleyet D.Genetic Algorithms and Neural networks:Opti-mizing Connections and Connectivity[J].Parallel Compution,1990,14:347~361.
[9]Kitano H.Designing Neural Networks Using Genetic Algo-rithms with Graph Generation System[J].Complex Sysmtem,1990,4:461~476.
[10]Jang J S R.,Sun C T,Mizutani E.Neuro-Fuzzy and SoftComputing[J].America:Prentice-Hall,1997.
[11]Glover F.Future paths for Integer Programming and Links toArtificial Intelligence[J].Computers and Operations Re-search,1986,5:533~549.
[12]李敏强,寇纪淞,林丹,等.遗传算法的基本理论与应用[M].北京:科学出版社,2003.
[13]李琼,李正文,魏野.同铁构造嘉陵江组储层裂缝非线性预测与分析研究[J].矿物岩石,2004,24(2):78~81.
[14]杨文采.神经网络在算法在地球物理反演中的应用[J].石油物探,1995,34(1):116~120.
[15]姚逢昌,甘利灯.地震反演的应用与限制[J].石油勘探与开发,2000,27(2):53~56.
[16]候安宁,何樵登.地震弹性波参数的非线性数值反演[J].石油地球物理勘探,1994,29(6):669~677.
[17]陈小宏,牟永光.二维地震资料波动方程非线性反演[J].地球物理学报,1996,39(3):401~408.
[18]杨磊,张向君,李幼铭.地震道非线性反演的参数反馈控制及效果[J].地球物理学报,1999,42(5):677~684.
[19]李庆忠著.走向精确勘探的道路[M].北京:石油工业出版社,1994.
[20]李琼,贺振华,黄德济,等.温压条件下孔洞模型超声波实验与结果分析[M].中国地球物理学会第十九届学术年会论文集,南京:南京师范大学出版社,2003.
[21]张宏兵,尚作萍,杨长春,等.波阻抗反演正则参数估计[J].地球物理学报,2005,48(1):181~188.
[2]王小平,曹立明.遗传算法———理论、应用与软件实现[M].西安:西安交通大学出版社,2002.Wang X P,Cao L M.Genetic algorithms—theory,application,software implement[M].Xi an:Xian Jiaotong UniversityPress,2002.
[3]刘勇,康立山,等.非数值并行计算(第2册)———遗传算法[M].北京:科学出版社,1995.Liu Y,Kang L S,et al.Non-numerical value parallel compu-ting(Ⅱ)———Genetic algorithms[M].Beijing:Science press,1995.
[4]焦李成.神经网络的应用和实现[M].西安:西安电子科技大学出版社,1993.Jiao L C,Neural network application and implement[M].Xian:XIDIAN university press,1993.
[5]A.И.加卢什金著.神经网络理论[M].阎平凡译.北京:清华大学出版社,2003.AlexanderИG.Neural network theory.Yan P F translating[M].Beijing:Tsinghua university press,2003.
[6]Jang J S R.ANFIS:Adaptive-Network-Based Fuzzy InferenceSystem[J].IEEE Transaction on System,Man and Cybernet-ics,1993,23(3):665~685.
[7]Goldberg D E.Genetic Algorithms in Search,Optimization,andmachine Learning[J].Reading,MA,Addison-Wesley,1989.
[8]Whitleyet D.Genetic Algorithms and Neural networks:Opti-mizing Connections and Connectivity[J].Parallel Compution,1990,14:347~361.
[9]Kitano H.Designing Neural Networks Using Genetic Algo-rithms with Graph Generation System[J].Complex Sysmtem,1990,4:461~476.
[10]Jang J S R.,Sun C T,Mizutani E.Neuro-Fuzzy and SoftComputing[J].America:Prentice-Hall,1997.
[11]Glover F.Future paths for Integer Programming and Links toArtificial Intelligence[J].Computers and Operations Re-search,1986,5:533~549.
[12]李敏强,寇纪淞,林丹,等.遗传算法的基本理论与应用[M].北京:科学出版社,2003.
[13]李琼,李正文,魏野.同铁构造嘉陵江组储层裂缝非线性预测与分析研究[J].矿物岩石,2004,24(2):78~81.
[14]杨文采.神经网络在算法在地球物理反演中的应用[J].石油物探,1995,34(1):116~120.
[15]姚逢昌,甘利灯.地震反演的应用与限制[J].石油勘探与开发,2000,27(2):53~56.
[16]候安宁,何樵登.地震弹性波参数的非线性数值反演[J].石油地球物理勘探,1994,29(6):669~677.
[17]陈小宏,牟永光.二维地震资料波动方程非线性反演[J].地球物理学报,1996,39(3):401~408.
[18]杨磊,张向君,李幼铭.地震道非线性反演的参数反馈控制及效果[J].地球物理学报,1999,42(5):677~684.
[19]李庆忠著.走向精确勘探的道路[M].北京:石油工业出版社,1994.
[20]李琼,贺振华,黄德济,等.温压条件下孔洞模型超声波实验与结果分析[M].中国地球物理学会第十九届学术年会论文集,南京:南京师范大学出版社,2003.
[21]张宏兵,尚作萍,杨长春,等.波阻抗反演正则参数估计[J].地球物理学报,2005,48(1):181~188.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心