【作者】 彭真明；

【导师】 肖慈(王旬)； 赵伟卫； 龚奇；

【作者基本信息】 成都理工学院 ， 地球探测与信息技术， 2001， 博士

【摘要】 地震反演是地震勘探的根本问题。由于大部分地球物理反演属于多极值的目标函数优化，使得基于线性或拟线性理论的反演方法往往导致反演结果陷入局部最优，造成反演结果的不可靠。本文结合前沿交叉学科的一些最新研究成果，研究提出了基于非线性优化理论的地震反演方法，并用于解决当前油气勘探开发中亟待解决而又未解决好的难题。该项研究不仅具有较高的学术价值，而且具有广阔的应用前景。 本文的主要研究内容和取得的成果有：①通过深入分析和研究优化问题中的遗传算法和模拟退火法各自的优劣势，提出了两种算法相互渗透、相互配合的退火遗传算法，从而改善了非线性全局寻优方法的搜索性能，提高了其在地球物理参数反演中的运算效率和反演精度。②针对神经网络结构设计不灵活、权值获取规则单一、易陷入局部最优、推广预测能力差等不足，提出了一种用遗传演化策略训练神经网络的方法，实现了网络连接权、阈值、网络最佳结构的自适应演化。③对地球物理反演中难度较大的剩余静校正量估计进行了深入研究，提出了混合优化自动剩余静校正方法，并在两个地区的实际资料处理中取得了较好的应用效果。④针对地震储层预测中存在的问题，提出了储层非线性反演与预测的方法技术，并成功应用于白云查干凹陷达尔其构造的三维地震反演和储层预测，为该区控制储量申报和有利目标区选择提供了准确的依据。⑤首次提出了一种井底以下地层压力随钻预测的新方法。该方法引用储层横向预测的基本思路，将测井与地震相结合，通过建立地震属性与测井声波之间的神经网络非线性映射模型，进而反推井底以下声波曲线并预测地层压力。该项技术为莺琼盆地DF1-1-11高温高压井的顺利钻进提供安全保障、降低钻井成本作出了重要的贡献。更多还原

【Abstract】 Seismic inversion plays a key role in geophysical prospecting. As most of geophysical inversions appear as objective function optimization with multi-extreme, which makes the conventional inversion methods based on linear or quasi-linear theory cause the solution to be optimized locally and unreliable. In this paper, the author presents a series of new methods for seismic inversion based on nonlinear optimization theory by making use of some latest research results of cross-subjects in leading edge, which are used to resolve the difficulties in petroleum exploration and development. This study has considerable academic value as well as wide prospects. In this paper, the main research works and the achievement obtained include the following contents: ?presents a hybrid optimization method combining genetic algorithm with simulated annealing by deeply analyzing and studying their advantage and disadvantage individually. The hybrid algorithm improved the searching performance in nonlinear global optimization, calculation efficiency and inversion accuracy in geophysical parameters inversion. ?To improve the low performance of existing network learning such as inflexibility of Topological structure designed ,simple regulation for weights obtained , being easy to be lost in local optimal solution and prediction with low performance, this paper presents a new way to train the neural network with genetic evolutionary strategy so as to make the linking weight, threshold , optimal structure adaptively evolve. ?By deeply study residual statics correction thought to be difficult in geophysics inversion, this paper presents a new approach to automatic residual statics using hybrid optimization, which brought satisfactory result when applied to real data of two areas ?To overcome the problem of existing reservoir prediction using seismic data, we developed the technology for nonlinear inversion and prediction of reservoir, which is successfully used in 3D seismic inversion and reservoir prediction for Darqi structure in Baiyunchagan depression and gives an accurate approach of submitting controlled reserves and selecting favorable targets for this area .?The first to present a new method for predicting formation pressure. ahead of bit. The method adopts the basic thought of lateral prediction of reservoir, combine logging and seismic data, builds the nonlinear mapping model of neural network between seismic attributes and sonic logging and then extrapolates sonic curves then predicts formation pressure ahead of bit. Great contribution has been done ~2 with this technique to successful drilling with safe warrant and low drilling cost for high temperature and pressure well DF 1-1-11 in Yingqiong basin.更多还原