摘要
针对全波形反演中Hessian矩阵庞大、共轭梯度法收敛速度慢的问题,提出了一种新的算法——快速共轭度梯度法(Fast Conjugate Gradient,FCG)。该方法通过引入新的变量对共轭梯度法进行改进,在加快收敛速度的同时使收敛更加稳定,额外的计算量是少许的点乘计算,计算量增加很少。将该法应用于频率域声波全波形反演中,并用简单的凹陷模型和抽稀的复杂Marmousi模型进行测试。测试结果表明:相对于传统的共轭梯度法,该方法能加快收敛速度,同时深部反演效果更好。
Hessian matrix in the full waveform inversion is huge and the convergence of the gradient method is slow. To solve the problem, we propose in this paper a new algorithm, fast conjugate gradient (FCG) method. The method introduces a new variable to transform the conjugate gradient method, which accelerates the convergence and makes convergence more stable. The method needs little more calculation of multiplication. The proposed method is applied to the acoustic full waveform inversion in the frequency domain and it is also tested in the simple depression model and the thinning complex Marmousi model. The tests show that the proposed method can accelerate convergence while the resolution of deep layer is better. © 2016, Editorial Department OIL GEOPHYSICAL PROSPECTING. All right reserved.
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2016年第4期730-737,4-5,共8页
Oil Geophysical Prospecting
基金
国家"973"项目(2013CB228604
2014CB239201)
国家油气重大专项(2016ZX05027004-001
2016ZX05002-005-09HZ)
国家自然科学基金-石油化工基金联合重点项目(U1562215)资助