基于学习字典的FAVO衰减参数反演方法研究

Compared with conventional AVO inversion, FAVO (Frequency-dependent AVO) inversion can obtain attenuation and dispersion attributes which are closely related to oil and gas in reservoirs. It is an important research direction in recent years. However, the FAVO inversion is an ill-posed problem, the existing regularization constraint method can hardly guarantee the stability and reliability of the inversion result. The regularization constraint has become a bottleneck restricting the application of the method. Focusing on the completeness and objectivity of the regularization constraints, this project will introduce the sparse representation and dictionary learning methods into the FAVO inversion. By learning, training and interpreting the logging data and rock physics data, the dictionary reflecting the internal structure of the parameters are adaptively constructed. Then, the FAVO inversion process is constrained by the sparse representation of the dictionary, to improve the FAVO inversion accuracy. The project includes two innovation research contents. The first is the prior Information extraction method based on the learning dictionary, and the second is the FAVO inversion method combined with the sparse representation based on learned dictionary. The achievement of this subject has important value and application potential for FAVO inversion.

与常规AVO反演相比，FAVO（Frequency-dependent AVO）反演能够得到与含油气性更加密切的地震波在储层中的吸收衰减和频散等信息，是近年来利用叠前地震反演进行油气储层预测的重要研究方向。然而，FAVO反演不适定问题严重，现有的先验约束方法很难保证反演结果的稳定性和可靠性，该问题已经成为制约FAVO反演实际应用的技术瓶颈。着眼于先验约束信息的完备性和客观性，本课题将基于学习字典的稀疏表征方法引入到FAVO反演，通过对测井数据和岩石物理数据的学习、训练和解译，自适应地构建反映参数内在结构的字典原子和表征体系，对反演过程进行约束和修正，提高FAVO反演精度。本项目主要包括两项创新研究内容，一是基于学习字典的FAVO储层多参数特征提取方法，二是结合字典稀疏表征的FAVO反演方法。本课题的研究对于充分发挥FAVO反演的理论优势，推动该技术的实际应用，具有重要意义。

FAVO（Frequency-dependent AVO）反演能得到与含油气性密切的储层吸收衰减和频散等信息，对储层预测具有重要意义。然而，FAVO反演不适定问题严重，现有的先验约束方法假设地下介质服从确定性的数据模型，很难保证先验信息对地下特征有效刻画。针对上述问题，本项目利用字典学习方法从已有测井数据、岩石物理数据和地震数据中学习参数先验信息，以数据驱动形式自适应地构建反映参数内在结构的字典原子和表征体系，通过稀疏表征形式对反演过程进行约束和修正，有效提高反演精度。本项目为数据驱动地震智能反演提供了新的思路，发展了基于联合字典的多参数稀疏表征约束反演方法、正交字典快速稀疏表征约束方法和多道联合稀疏表征约束反演方法，在地震反演提高稳定性，降低多解性和提高计算效率等方法形成了多项创新。项目共发表期刊论文13篇，申请或授权专利17项，发表会议论文9篇。此外项目资助下发展了基于稀疏表征方法地震反演软件模块，在塔里木油田、西南油气田等工区进行了应用，取得了显著效果。