预处理DYHS混合共轭梯度时域波形反演算法

预处理DYHS混合共轭梯度时域波形反演算法 A Preconditioned DYHS Hybrid Conjugate Gradient Method for Time-Domain Full-Waveform Inversion 杨艳 YANG Yan 1 first-author 谢思琦 XIE Siqi 1 张东 ZHANG Dong 1 秦前清 QIN Qianqing 2 胡英 HU Ying 3 武汉大学物理科学与技术学院 武汉大学物理科学与技术学院 School of Physics and Technology,Wuhan University School of Physics and Technology,Wuhan University 武汉大学测绘遥感信息工程国家重点实验室 武汉大学测绘遥感信息工程国家重点实验室 State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University 中国石油勘探开发研究院物探技术研究所 中国石油勘探开发研究院物探技术研究所 Department of Geophysical Technology,Research Institute of Petroleum Exploration and Development,PetroChina Department of Geophysical Technology,Research Institute of Petroleum Exploration and Development,PetroChina 为了提高梯度类算法的反演效果及其收敛性,本文对DYHS混合共轭梯度法进行了研究,采用近似黑塞矩阵的主对角元及正则化技术对梯度进行预处理,并对该共轭梯度法进行了改进.以变化层状阶梯模型为例,分别实现了传统的梯度类方法以及预处理后的DYHS混合共轭梯度法,反演结果验证了改进后的反演算法能很好地提高反演的收敛性,使模型结果更接近于理论模型. DYHS hybrid conjugate gradient method is studied in this paper. In order to improve the inversion performance of the gradient-based method and accelerate the convergence,the diagonal elements of approximate Hessian matrix combing with the regularization technique is applied in preprocessing the gradient for adapting the conjugate gradient method. The traditional gradient method and the preconditioned DYHS hybrid conjugate gradient method are all implemented on the varying layered stair-step model. The results show that the improved method not only accelerates the convergence,but also makes the inversion results closer to the theoretic model. 混合共轭梯度法 预处理 时域波形反演 hybrid conjugate gradient method preconditioned time-domain waveform inversion 10.14188/j.1671-8836.2015.05.005 O224 国家科技重大专项（2011ZX05003-003）基金资助 2015-05-01 2021-03-31 5 61