叠前多属性概率神经网络反演储层孔隙度
|
摘要
近十年来非线性反演方法(如人工神经网络、遗传算法)在地球物理解释中,得到越来越多的应用,但人工神经网络反演目前通常只采用叠后波阻抗反演结果和叠后地震属性(如振幅、频率和相位)进行预测,而忽略了地震叠前道集中包含的地层信息。这里通过叠前地震反演获得纵波、横波阻抗和密度信息,结合叠前地震属性,综合应用PNN神经网络方法来反演地层孔隙度参数。其过程包括:①提取叠前地震属性和叠前反演纵波、横波阻抗和密度参数;②分析孔隙度和各类叠前属性和叠前弹性参数的相关程度,确定出与孔隙度关系密切的主要参数;③综合叠前反演弹性参数和叠前属性等参数,应用神经网络分析方法反演得出孔隙度体。该方法克服了由于砂泥岩波阻抗重叠造成的叠后波阻抗反演储层预测存在多解性的问题,反演孔隙度体提高了储层识别精度,储层预测和钻井结果一致,符合实际地质规律,证明本方法正确有效。
Non-linear seismic inversion is always difficult problem.Some no-linear inversion method such as neural network,are being applied in the seismic interpreting.But they only base on post-stack seismic attributes.The paper developed a new method,which inverted porosity by method of PNN network basing on pre-stack inversion result and pre-stack seismic attribute.The process includes 3 steps.Firstly,some pre-stack attributes are extracted and the elastic parameters are inverted by pre-stack inversion.Secondly,analysis of the log data is carried out to acquire the relationship between porosity and pre-stack information.At last,with pre-stack inversion results and AVO attributes,PNN model is applied to get porosity volume.The technology eliminates the uncertainty of post-stack impendence inversion.The inverted porosity volume increases reservoir precision and is consistent with geologic result.
Non-linear seismic inversion is always difficult problem.Some no-linear inversion method such as neural network,are being applied in the seismic interpreting.But they only base on post-stack seismic attributes.The paper developed a new method,which inverted porosity by method of PNN network basing on pre-stack inversion result and pre-stack seismic attribute.The process includes 3 steps.Firstly,some pre-stack attributes are extracted and the elastic parameters are inverted by pre-stack inversion.Secondly,analysis of the log data is carried out to acquire the relationship between porosity and pre-stack information.At last,with pre-stack inversion results and AVO attributes,PNN model is applied to get porosity volume.The technology eliminates the uncertainty of post-stack impendence inversion.The inverted porosity volume increases reservoir precision and is consistent with geologic result.
引文
[1]张赛民,周竹生,陈灵君,等.应用一种改进BP神经网络算法预测密度曲线[J].物探化探计算技术,2007,29(6):497.
[2]马英杰,周蓉生.神经网络方法在岩性识别中的应用[J].物探化探计算技术,2004,26(3):220.
[3]李海燕,彭仕宓.应用遗传神经网络研究低渗透储层成岩储集相——以胜利渤南油田三区沙河街组为例[J].石油与天然气地质,2006,27(1):111.
[4]张绍红.概率神经网络技术在非均质地层岩性反演中的应用[J].石油学报,2008,29(4):549.
[5]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京:清华大学出版社,2000.
[6]Tsai C.Y.An Iterative Feature Reduction Algorithmfor Probabilistic Neural Networks[J].Omega,2000,28:513.
[7]邹才能,张颖.油气勘探开发实用地震新技术[M].北京:石油工业出版社,2002:315.
[8]戴晓峰.Application of joint elastic impedance inver-sion in the GD Oilfield[J].Applied Geophysics,2006,3(1):37.
[9]管路平.塔河油田南部缝孔型碳酸盐岩储层叠前地震预测[J].石油与天然气地质,2008,29(5):654.
[10]于建国,姜秀清.地震属性优化在储层预测中的应用[J].石油与天然气地质,2003,24(3):291.
[11]杨占龙.地震属性分析与岩性油气藏勘探[J].石油物探,2007,46(2):131.
[12]印兴耀.地震属性优化方法综述[J].石油地球物理勘探,2005,40(4):482.
[13]黄飞.PNN及其在储层物性参数预测中的应用[J].物探化探计算技术,2002,25(1):78.
[2]马英杰,周蓉生.神经网络方法在岩性识别中的应用[J].物探化探计算技术,2004,26(3):220.
[3]李海燕,彭仕宓.应用遗传神经网络研究低渗透储层成岩储集相——以胜利渤南油田三区沙河街组为例[J].石油与天然气地质,2006,27(1):111.
[4]张绍红.概率神经网络技术在非均质地层岩性反演中的应用[J].石油学报,2008,29(4):549.
[5]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京:清华大学出版社,2000.
[6]Tsai C.Y.An Iterative Feature Reduction Algorithmfor Probabilistic Neural Networks[J].Omega,2000,28:513.
[7]邹才能,张颖.油气勘探开发实用地震新技术[M].北京:石油工业出版社,2002:315.
[8]戴晓峰.Application of joint elastic impedance inver-sion in the GD Oilfield[J].Applied Geophysics,2006,3(1):37.
[9]管路平.塔河油田南部缝孔型碳酸盐岩储层叠前地震预测[J].石油与天然气地质,2008,29(5):654.
[10]于建国,姜秀清.地震属性优化在储层预测中的应用[J].石油与天然气地质,2003,24(3):291.
[11]杨占龙.地震属性分析与岩性油气藏勘探[J].石油物探,2007,46(2):131.
[12]印兴耀.地震属性优化方法综述[J].石油地球物理勘探,2005,40(4):482.
[13]黄飞.PNN及其在储层物性参数预测中的应用[J].物探化探计算技术,2002,25(1):78.
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