【作者】 余海龙；

【导师】 徐世浙；

【作者基本信息】 浙江大学 ， 地球探测与信息技术， 2009， 博士

【摘要】 岩石的密度、磁性与岩石类型、区域构造有着密切联系,研究岩石的密度和磁性分布对位场异常的解释、研究大地构造分区及矿产资源调查具有重要的意义。如何根据位场异常计算地球内部岩石的密度、磁化率或磁化强度分布,称之为位场异常的物性反演。但由于位场的叠加和等效性等原因,物性反演问题一直是国内外研究的热点和难点问题。目前位场物性反演的方法大致可分为最优化选择法和非线性反演方法两大类,具体的算法有广义最小二乘法(Gauss法)、最速下降法、阻尼最小二乘法(Marguardt法)、奇异值分解法、脊回法、人工神经网络法BP算法、遗传算法、模拟退火算法和基于Radon变换、多尺度边缘的算法等。这些算法的基本思路是,将地下场源区域一次性划分成若干形态已知、密度或磁性未知的小棱柱体单元,首先给定密度或磁性等物性参量一个初始值,然后通过各种反演算法和各种约束条件求得参量的改变值,最终求得合理的物性参量,使得模型的理论正演异常与实测异常之差的L1或L2范数等最小。目前已有的这些算法,在理论上是合理、可行的,但在实际应用中却存在一些问题。其中,最优化选择法的主要问题,一是需要建立大型的线性代数方程组,计算时间很长;二是方程组病态严重,解的稳定性很差。非线性反演方法的主要问题是,反演数据量较大时,存在超常规的计算量、存储量及收敛速度慢的问题。总体来说,由于目前计算机的内存和速度的限制,在对位场资料的物性反演,尤其是大面积资料的三维物性反演时,已有的反演方法都很难奏效。针对现有的问题,本文在系统研究了位场分离方法、位场向下延拓方法和频率域层源位场异常快速反演方法的基础上,提出了一种基于位场分离与延拓的三维物性快速反演方法。方法的基本思路为,首先运用位场分离的插值切割法对平面上的场进行不同深度层源的分离,得到各深度层源在地面引起的异常;然后应用大深度的位场迭代法向下延拓技术将各深度层源的地面异常延拓到各深度层的顶部;最后,根据各深度层顶部的异常利用基于棱柱体组合的频率域反演方法反演出各个深度层源的密度或磁性。新反演方法不需要对场源进行一次性剖分和解线性代数方程组,避开了制约三维反演实用化的超大内存需求、解稳定性差和收敛慢引起的特别冗长的计算时间的瓶颈问题。通过新方法对新疆色尔特能地区、普光气田和东海及邻区重力异常的视密度反演和对霍邱铁矿磁异常的视磁性反演,表明该方法具有精度高、稳定性强和速度快的优点,适用于大面积位场资料的三维物性反演,具有一定的理论和实际意义。运用Fortran语言、c语言和C++Builder编译环境,结合动态链接库技术和多线程技术对新方法编制了Windows系统下的可视化软件,方便了方法的应用和推广。更多还原

【Abstract】 The density and magnetization of rock are closely related to the lithology and tectonics. Therefore, the study on the density and magnetization of rock is very significant for the interpretation of potential anomaly data, tectonic unit division and mineral resources investigation. The calculating of rock density and magnitization according to potential anomaly is called physical properties inversion of potential anomaly. Because of potential superposition and equivalence principle, the research of physical properties inversion has been an interested and difficult problem at home and abroad.At present, there are two main methods to inverse the physical properties of potential anomaly, such as optimization method and non-linear method. The algorithms of these two methods include Gauss algorithm, the steepest descent algorithm, Marquardt algorithm, singular value decomposition algorithm, ridge regression algorithm, the Back Propagation algorithm of artifical neural networks, genetic algorithm method and simulated annealing algorithm et al. The basic thinking of present methods is to divide the subsurround field sources region into many rectangular blocks with known shape and unknown density or magnetization. At first, initial values of density or magnetization are given. Then, the changed values of density or magnetization are needed to be gained. Finaly, the reasonable physical properties values that make the L1 norm or L2 norm of the difference between theory forward anomaly data and observed data to be least are obtained.Even though the present methods are reasonable and feasible in theory, there are still many problems in practice. There are mainly two problems about optimization method. Firstly, the establishment of large linear algebraic equation is needed, it consumes large computer memory and long computation time. Secondly, the equation is ill-conditioned and solution is unstable. The main problem of non-linear inversion methods is the long calculating time cosuming and large memory occupying when the quantities of inverse data are large. Overall the present inversion methods are ineffective during inversion of potential data, especially 3-D physical properties inversion in the large area due to the restriction of memory and computing velocity of computer.This paper presents a rapid method for 3D physical properties inversion based on separation and continuation of potential field. The new method is carried out in the following three steps. Firstly, potential feilds corresponding to strata of different depth on horizontal groud are gained through separating the potential fields on the plane by a cutting method. Secondly, the fields corresponding to strata of different depth on horizontal groud are reduced to a plane corresponding to the top surface of different depth strata by using downward continuation. Finally, according to the field on the top surface, inversion is conducted to get structure of apparent density or magnetization in frequency. The inversion technique characterized by faster computing speed, it does not need to divide the subsurround field sources region into many rectangular blocks in one time and does not need to slove linear algebraic equations as well. So, this technique can overcome the requirement of large computer memory, unstable solution and computation time bottleneck which hinders the application of 3Dinversion to practice.Several cases, including apparent density inversion of gravity anomolies in Xinjiang, Puguang gas field and the East China Sea and its adjacent area as well apparent magnetization inversion of magnetic anomaly in Huoqiu iron mine area, are presented to demonstrate the effect of applying this method. This method is suitable for 3D inversion of potential field in large area and has some theorical and practical significance. A visual software in Windows system is designed with Fortran language, C language, DLL technology and multithread technology based C++ Builder 6.0, which provides convenience for the use and popularization of new method.更多还原