走时反演RAYINVR算法试验及其在镇康—泸西剖面地壳结构重建中的应用

发布时间：2018-03-29 10:24

本文选题：地震走时反演　切入点：算法试验　出处：《中国地震局地球物理研究所》2017年硕士论文

【摘要】：本文详细研究了Zelt和Smith在1992年提出的射线反演Rayinvr算法。模型参数化灵活设置速度和界面节点的个数和位置,使不规则四边形网格大小的划分与地下介质被数据分辨的状况相一致,很大程度上避免了过参数化引起的反演假象和欠参数化引起的不能分辨细结构的缺陷。层边界光滑技术的应用避免了模型参数化边界转折点造成的人为的射线聚焦和焦散,进而在地表接收点引起非客观的地震射线影区和三重震相的现象。四边形内速度采用双线性内插,使得内部射线追踪存在解析解。到达观测点的射线追踪采用试射法,大大提高了算法的效率。反演使用阻尼最小二乘算法,提高了算法的稳定性。结果在追踪到射线数、走时残差和阻尼因子之间取折中。对影响该算法的因素:阻尼因子选取、初始模型选取、走时随机扰动和数据不确定性进行了试验,测试了多种因素对反演的影响。阻尼因子是影响算法效率和结果分辨率的关键因素。在实际资料处理过程中,取大的阻尼值将使反演迭代加长,降低算法的效率,而且大大降低了结果分辨率;阻尼值取的较小,可能使反演无法进行。为了提高计算效率和得到分辨较佳的最终结果,在每次迭代时都需要尝试不同的阻尼值,这是非常耗时间的。对此,我们对阻尼因子选取进行改进,并做了测试试验。我们做法是在初始迭代中给阻尼因子赋一个比较大的值,以后在迭代的每一步中通过给该值乘以一个小于1的数,逐渐减小阻尼因子,增大收敛速度和提高模型参数分辨率。测试表明,完成反演只需几次选取这个小于1的乘子就可以了。这种改进大大提高了计算效率。Rayinvr地震走时反演算法是一种线性化反演算法,从十分不同的初始模型开始反演有可能收敛次级极值点,搜索不到全局最优解。对此,我们进行了另外两个试验,以获得算法对观测数据不确定性和不同的初始模型的敏感程度。结果表明,在对初始模型选取时应尽可能的使用先验信息,能有效地避免反演不能进行和追踪到的射线数过少等问题;而观测数据的获取也应该尽量准确。通过对镇康-泸西人工地震剖面的资料处理,重建了滇南中部高分辨的二维地壳上地幔速度结构。反演中使用参数和初始模型基于试验内容进行选取,结果与正演相比有所改善。地壳结构主要以沉积层、C1、C2、C3、莫霍面分层,每层都出现了明显的横向速度不均匀性,以桩号400km处红河断裂带为界,西侧速度偏低,而东侧速度偏高,莫霍面深度明显增厚。在小江断裂带出现明显低速,往东进入华南大陆为稳定陆块。
[Abstract]:In this paper, the authors study in detail the ray inversion Rayinvr algorithm proposed by Zelt and Smith in 1992. The model is parameterized to set the velocity and the number and position of interface nodes. The partition of irregular quadrilateral mesh size is consistent with the resolution of underground media by data. To a great extent, the defects of inversion illusion caused by over-parameterization and indiscernibility of fine structure caused by under-parameterization are avoided. The application of layer boundary smoothing technique avoids the artificial ray focusing and defocusing caused by the turning point of model parameterized boundary. The velocity in the quadrilateral is bilinear interpolated so that there is an analytical solution to the internal ray tracing. The ray tracing to the observation point is carried out by test shooting method, which leads to the phenomenon of non-objective seismic ray area and triple seismic phase at the receiving point of the earth's surface, and the velocity of the quadrilateral is bilinear interpolated. The efficiency of the algorithm is greatly improved. The stability of the algorithm is improved by using the damped least square algorithm. Results the tradeoff between tracing the number of rays, the travel time residuals and the damping factor is taken. The selection of initial model, the random disturbance of travel time and the uncertainty of data are tested, and the influence of many factors on inversion is tested. Damping factor is the key factor to affect the efficiency and resolution of the algorithm. Taking a large damping value will make the inversion iteration longer, reduce the efficiency of the algorithm, and greatly reduce the resolution of the result. A smaller damping value may make the inversion impossible. In order to improve the computational efficiency and obtain a better resolution of the final results, We need to try different damping values at each iteration, which is very time-consuming. For this, we improve the damping factor selection and do a test experiment. We assign a larger value to the damping factor in the initial iteration. In each step of the iteration, by multiplying the value by a number less than 1, the damping factor is gradually reduced, the convergence rate is increased and the resolution of the model parameters is improved. It takes only a few times to select the multiplier less than 1 to complete the inversion. This improvement greatly improves the calculation efficiency. Rayinvr seismic walk time inversion algorithm is a linearized inversion algorithm. From the very different initial model, the possible convergence of the secondary extremum point is inversed, and the global optimal solution can not be found. In this case, we have carried out two other experiments. In order to obtain the sensitivity of the algorithm to the uncertainty of observation data and the different initial models, the results show that priori information should be used as far as possible in selecting the initial model. It can effectively avoid the problem that the inversion can not be carried out and the number of rays that can be traced is too small, and the acquisition of observation data should be as accurate as possible. By processing the data of Zhenkang Luxi artificial seismic profile, The high resolution 2-D crust and upper mantle velocity structure in the central part of southern Yunnan was reconstructed. The parameters and initial model were selected based on the experimental contents in the inversion, and the results were improved compared with the forward model. The crustal structure was mainly composed of sedimentary layer C _ (1) C _ (2) C _ (2) and Moho stratification. There is obvious lateral velocity inhomogeneity in each layer, with the Honghe fault zone at pile number 400km as the boundary, the velocity on the west side is low, and the velocity on the east side is on the high side, and the Moho depth is obviously thickened, and in the Xiaojiang fault zone there is an obvious low velocity. Eastward access to the South China continent is a stable continent.
【学位授予单位】：中国地震局地球物理研究所
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P631.4

【参考文献】