W地区储层孔隙度地震预测技术及应用研究
|
摘要
在W地区,储层孔隙度预测技术是将支持向量机方法与分频法相结合构成一种新的地震非线性预测技术。支持向量机是一种基于统计理论的新型机器学习算法,它是建立输入与输出的一个隐形的映射关系,在我们的问题中,应用测井数据作为支持向量机的输入向量,通过支持向量机优选参数,对实际样本数据构造核函数,实现全局最优解。分频法采用短时快速傅利叶变换(SFFT)将地震数据进行时频转换,形成高分辨的分频数据体,作为独立的输入数据体,以提高储层孔隙度反演的分辨率和精度。因此,构成的这种联合预测技术,充分发挥了支持向量机和分频法的优势,这种全新的储层孔隙度预测技术适应储层地质参数高维非线性特点。依据建立的储层孔隙孔隙度预测技术,利用W地区实际地震数据形成的分频数据体和测井数据,对朱海组储层孔隙度进行了预测,获得了高分辨率和高精度储层孔隙度剖面和储层孔隙度数据体。在W地区,预测的朱海组储层孔隙度与测井孔隙度具有良好的一致性,其相关系数值为(0.816～0.945),平均值为0.882,因此,预测的储层孔隙度具有较高的可信度和准确率。在W地区成功应用表明,其方法技术具有通用性和较好的推广性,它的适用面广,预测效果好,易于实现,它将油气储层孔隙度预测推向一个新的发展水平。
In the W district,a new nonlinear seismic prediction method combined with the SVM method and the frequency division method are used in reservoir porosity prediction.As a new machine learning algorithm based on statistical theory,the SVM built an invisible mapping relationship between input and output.In fact,the log data are applied as the input vector of the SVM.Then,optimized parameters are picked by the SVM,the kernel function for sampled data is constructed,and the global optimum is achieved.While,the frequency division method uses the Short-Time Fast Fourier Transform(SFFT) to seismic data to recognize frequency transformation,from which frequency divides data volume with high resolution can be got and input as individual data volume to improve the solution and accuracy of reservoir porosity inversion.The combined prediction method is of the advantage of two methods mentioned above,and it works well in the reservoir of high nonlinear features.In this paper,the reservoir porosity prediction technique is applied to the frequency divided data and the log data,based on actual seismic data of the W district,to predict the porosity of Zhuhai Formation,and obtain its reservoir porosity profile and data volume of high solution and accuracy.As result,the prediction porosity of Zhuhai Formation was consistent with the log porosity,with their correlation coefficient of 0.816~0.945,and with average value of 0.882,demonstrating the high reliability and accuracy of this prediction method.The successful application of this technique in the W district proves that it is of great generality and can be extended to more cases with good prediction result.
In the W district,a new nonlinear seismic prediction method combined with the SVM method and the frequency division method are used in reservoir porosity prediction.As a new machine learning algorithm based on statistical theory,the SVM built an invisible mapping relationship between input and output.In fact,the log data are applied as the input vector of the SVM.Then,optimized parameters are picked by the SVM,the kernel function for sampled data is constructed,and the global optimum is achieved.While,the frequency division method uses the Short-Time Fast Fourier Transform(SFFT) to seismic data to recognize frequency transformation,from which frequency divides data volume with high resolution can be got and input as individual data volume to improve the solution and accuracy of reservoir porosity inversion.The combined prediction method is of the advantage of two methods mentioned above,and it works well in the reservoir of high nonlinear features.In this paper,the reservoir porosity prediction technique is applied to the frequency divided data and the log data,based on actual seismic data of the W district,to predict the porosity of Zhuhai Formation,and obtain its reservoir porosity profile and data volume of high solution and accuracy.As result,the prediction porosity of Zhuhai Formation was consistent with the log porosity,with their correlation coefficient of 0.816~0.945,and with average value of 0.882,demonstrating the high reliability and accuracy of this prediction method.The successful application of this technique in the W district proves that it is of great generality and can be extended to more cases with good prediction result.
引文
[1]Boser B,Guyon I,Vapnik V.A training algorithm for optimal marginclassifiers.Proceedings of the 5th Annual ACM Workshop on Compu-tational Learning Theory[J].ACM Press,1992:144-152.
[2]Vapnik V.SVM method of estimating density,conditional probability,and conditional density[C].IEEE International Symposium on Circuitsand Systems,v 2,2000,pII-749-752.
[3]Vapnik V N.统计学习理论的本质[M].张学工译.北京:清华大学出版,2000.
[4]Loosli,Galic,Canu,Stéphane;Vishwanathan,S V N;Smola,Alexander J;Chattopadhyay,Manojit.Simple and fast support vector ma-chine(SVM)toolbox[J].Revue d′Intelligence Artificielle,2005,v19,n 4-5:741-767.
[5]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1):32-42.
[6]杨晓伟,路节,张广全.一种高效的最小二乘支持向量机分类器剪枝算法[J].计算机研究与发展,2007,44(7):1128-1136.
[7]聂玲玲,童凯军,侯东梅,等.利用支持向量机方法求取缝洞型油气藏地层系数[J].中国海上油气,2011,23:93-96.
[8]吴芳,王卫东,张永志,等.基于小波分析和最小二乘支持向量机的中国大陆地震震级预测研究[J].地震,2010,30:4-60.
[9]李勇.地震优化非线性反演方法及应用研究[J].矿物岩石,2004,24(1):101-104.
[10]李勇,王绪本,徐炳高,等.基于混合模糊神经网络储层裂缝地震反演研究[J].矿物岩石,2006,26:116-119.
[11]李培明.利用地震属性二次规划反演孔隙度参数[J].石油地球物理勘探,2003,38(2):182-184.
[12]王永刚.综合地球物理信息的孔隙度预测方法及应用效果分析[J].石油地球物理勘探,2001,36(6):707-715.
[13]李正文,杨谦.地震岩性预测方法及应用[J].矿物岩石,1996,16:81-85.
[14]黄绪德.油气预测与油气藏描述——地震勘探直接找油气[M].南京:江苏科学技术出版社,2003.
[15]陈欢庆,朱筱敏,刘长利,等.A盆地深区L组海相地层层序地层特征[J].矿物岩石,2009,29(1):114-120.
[16]张贤达.现代信号处理[M].北京:清华大学出版社,2002.
[2]Vapnik V.SVM method of estimating density,conditional probability,and conditional density[C].IEEE International Symposium on Circuitsand Systems,v 2,2000,pII-749-752.
[3]Vapnik V N.统计学习理论的本质[M].张学工译.北京:清华大学出版,2000.
[4]Loosli,Galic,Canu,Stéphane;Vishwanathan,S V N;Smola,Alexander J;Chattopadhyay,Manojit.Simple and fast support vector ma-chine(SVM)toolbox[J].Revue d′Intelligence Artificielle,2005,v19,n 4-5:741-767.
[5]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1):32-42.
[6]杨晓伟,路节,张广全.一种高效的最小二乘支持向量机分类器剪枝算法[J].计算机研究与发展,2007,44(7):1128-1136.
[7]聂玲玲,童凯军,侯东梅,等.利用支持向量机方法求取缝洞型油气藏地层系数[J].中国海上油气,2011,23:93-96.
[8]吴芳,王卫东,张永志,等.基于小波分析和最小二乘支持向量机的中国大陆地震震级预测研究[J].地震,2010,30:4-60.
[9]李勇.地震优化非线性反演方法及应用研究[J].矿物岩石,2004,24(1):101-104.
[10]李勇,王绪本,徐炳高,等.基于混合模糊神经网络储层裂缝地震反演研究[J].矿物岩石,2006,26:116-119.
[11]李培明.利用地震属性二次规划反演孔隙度参数[J].石油地球物理勘探,2003,38(2):182-184.
[12]王永刚.综合地球物理信息的孔隙度预测方法及应用效果分析[J].石油地球物理勘探,2001,36(6):707-715.
[13]李正文,杨谦.地震岩性预测方法及应用[J].矿物岩石,1996,16:81-85.
[14]黄绪德.油气预测与油气藏描述——地震勘探直接找油气[M].南京:江苏科学技术出版社,2003.
[15]陈欢庆,朱筱敏,刘长利,等.A盆地深区L组海相地层层序地层特征[J].矿物岩石,2009,29(1):114-120.
[16]张贤达.现代信号处理[M].北京:清华大学出版社,2002.
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