Extraction of ground structure for seismic zoning using inverse analysis

从多个强震记录中还原大面积地面结构具有重要意义。提出了一种改进的反射/透射矩阵法和静态AL法来考虑土层边界上产生的散射波。应用该方法进行的算例分析表明，散射波对地面响应特性有两种影响。一是在频响函数上给出了不能用一维分析表示的其他峰值，二是与频率相关阻尼的效果相同。接下来，我们开发了一种方法，通过波形反演方案，利用在地表观测到的地震图来确定地面结构的边界形状。本文研究了如何在二维SH问题的情况下，从表面观测记录确定边界形状的数值方法。为简单起见，我们只参数化边界形状。当给出一个合适的模型时，即使只给出几个地面台站的波形数据，也能准确地估计出边界的形状。本研究还探讨了利用神经网络方法识别地面结构系统动力特性的潜力。由于神经网络的自学习特性，其识别的动态特性受教学信号中所含噪声水平的强烈影响。利用卡门滤波技术，提出了一种识别教学信号中抗污染噪声的结构系统动态特性的方法。

It is very impotant to reduce ground structure of wide area from several strong motion records. A modified Reflection/Transmission Matrices method and stcastic AL method are proposed to consider scattered wave induced on the boundary between soil layrs. The case studies using this method show that the scattered wave has two effects on the characteristics of the ground response. One is to give other peaks on frequency response function which can not be represented by one dimensional analysis and the other is the same effect with the frequency dependent damping.Next we develop a method to determine the boundary shape of ground structures using seismograms observed on the surface through a waveform inversion scheme. This study examines numerically how to determine the boundary shape from observed records on the surface in the case of the two dimensional SH problem. For simplicity, we parameterize only the boundary shape. When an appropriate model is given, we can evaluate the boundary shape accurately even though the waveform data at only a few surface stations are given.This study also explores the potential of using neural network approaches to identify the dynamic characteristics of ground structural system. Because of self-learning nature of neural network the identified dynamic characteristics are strongly affected by the level of noise cotained in the teaching signals. Using the Karman filtering technique, a method to identify the dynamic characteristics of structural system proof against contaminating noise in teaching signals has been developed.