摘要
地震反演可以通过地表观测数据来推测地下介质模型参数,因此地震反演技术自提出至今都是储层预测最常用的手段之一.不动点地震反演方法(FPSI)通过将求解目标泛函转换为寻找目标泛函不动点来获得准确与高分辨率的反演结果.由于FPSI对地震信号的振幅变化具有较高的敏感性,因此即使地震剖面存在部分弱反射区域,利用FPSI仍然可以获得较为准确的反演结果.然而FPSI存在抗噪能力不足以及多解性强等问题.为提高FPSI的抗噪能力及实用性,本研究将随机反演框架引入至不动点反演理论,提出随机不动点地震反演方法(SFPSI).此外,利用纵波阻抗的初始模型计算得到初始反射系数,并将初始反射系数与初始模型约束引入反演框架及目标泛函,有效提高了反演方法的鲁棒性及计算效率.具体地,首先对以待反演参数初始模型为均值的高斯分布进行随机抽样并利用不动点理论判定抽样结果是否被接受.在此基础上对反演参数初始模型进行更新,并利用更新后的反演结果进行下一轮抽样与判定,直至残差小于阈值或迭代次数大于设定值.因此,SFPSI通过反复执行"随机过程"与"不动点修正"来使得初始模型逐渐逼近真实解,并最终收敛于目标泛函的不动点.由于本方法具有一定的随机性,因此需要进行多次反演并以多次反演结果的均值作为最终反演结果.最后,模型以及实际资料测试结果表明本方法有效的提高了FPSI抗噪能力,且本方法的反演结果较FPSI的反演结果具有更高的地层分辨率,并保持了良好的横向连续性.
The model parameters of the subsurface can be estimated from the observations by the seismic inversion methods.Thus,the seismic inversion is most utilized for reservoir prediction since it was proposed.The Fixed-Point Seismic Inversion(FPSI)can estimate the inverted results by transforming the problem of solving objective function to the problem of finding the fixed-point of the objective function.The accurate inversion results can also be obtained even in the weak reflection areas in seismic profile since the FPSI is sensitive to the small changes of the signal.However,the FPSI suffers from high noise-sensitivity and multi-solution.The Stochastic Fixed-Point Seismic Inversion(SFPSI)method is thus proposed in this study to alleviate these issues and improve the practicality of the FPSI.The initial model constraint is involved in this method to improve the lateral continuity and alleviate the multi-solution.Firstly,the initial model of the reflectivity can be calculated by the prior model of the P-wave impedance,and the initial models of the reflectivity should be sampled randomly from the Gaussian distribution and the sampled signal is judged whether it should be accepted or rejected by the theory of the fixed-point.And we should update inversion results no matter whether the sampled signal is accepted or rejected.The updated signal can then be utilized as a new signal to do the next iteration.And the loop ends when the error less than the threshold or the iterations reached the set value.Therefore,the SFPSI performs"stochastic process"and"fixed point correction"repeatedly to make the initial inversion results approach the real solutions gradually.And the solutions converge to the fixed-point of the objective function.Since the proposed method is stochastic,several times of inversion parameters should thus be obtained and the inverted results can be expressed as the average of the multiple inversion results.Finally,the model and field data examples demonstrate that the inversion results of our method has higher resolution than that of the FPSI.And the anti-noise ability of our method is improved sharply while the inversion results of our method has well lateral continuity.
作者
裴松
印兴耀
魏玮
李坤
PEI Song;YIN XingYao;WEI Wei;LI Kun(China University of Petroleum School of Geosciences,Qingdao 266580,China;Laboratory for Marine Mineral Resources,Qingdao National Laboratory for Marine Science and Technology,Qingdao 266580,China;Shandong Provincial Key Laboratory of Deep Oil and Gas,Qingdao 266580,China;Research Institute of Exploration and Development,Southwest Oil and Gas Field Company,PetroChina,Chengdu 610000,China)
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2023年第8期3413-3426,共14页
Chinese Journal of Geophysics
基金
国家自然科学基金(42030103,42004092)
中国博士后科学基金(2020M672170)联合资助。
关键词
不动点
随机
地震反演
计算效率
抗噪性
分辨率
Fixed-point
Stochastic
Seismic inversion
Computational efficiency
Anti-noise ability
Resolution
