【作者】 师黎静；

【导师】 陶夏新；

【作者基本信息】 中国地震局工程力学研究所 ， 防灾减灾工程及防护工程， 2007， 博士

【摘要】 场地条件对地震动和震害有十分显著的影响,相关研究一直是工程地震和地震学领域的重要内容。地震动研究中的众多疑难问题归根结底是由于对场地条件缺乏了解造成的。利用地脉动观测推断工程场地近地表三维速度结构,会为地球科学这一分支的发展提供有力的技术手段。本文利用地脉动方法进行场地近地表三维速度结构探测与建模成像研究,主要取得了如下成果。1.发展了一套适用于不同工况下的利用瑞利波频散曲线推算剪切波速度结构的免疫启发遗传反演策略,以浮点数编码、轮盘赌选择、实数均匀交换与变异算子相互匹配为基本遗传反演框架,以免疫启发为局部搜索强化策略,免重复计算为加速策略,多次运行,经模拟退火算法局部化后取最优为最终结果。借助虚拟反演思路,通过对各种遗传算子和编码方式不同匹配方式的比较研究,指出对于频散曲线反演浮点数编码与轮盘赌选择的匹配方式离线性能最好,提出利用浮点数编码、轮盘赌选择、浮点数均匀交换与变异算子匹配组成基本遗传算法框架。本文算法在基本遗传框架基础上施加了免疫启发策略和免重复计算加速策略。前者充分利用最佳个体的信息加速进化进程,通过对每代的最佳个体施加一服从标准正态分布的随机数来加强对邻近区域的局部搜索,同时通过标准差的调整也兼顾了对邻近区域以外区域的搜索,将局部搜索和全局搜索有机地结合起来,最大限度地降低了对遗传算法自身进化进程的干扰;通过施加免重复计算策略,大大减少了正演计算次数,节约了计算成本,提高了反演效率。通过对复杂工况的分析,指出了等厚薄层策略的不足,设计了固定层数、以土层厚度和波速为待反演参数的策略。同时,建议了多次运行,对每次运行结果继续施加模拟退火算法使其至少达到局部最优,最后取最优解的作法,避免了多次平均法的缺陷,提高了反演结果的稳定性和精度,降低了非唯一性。2.提出了一种多个场地联合、地脉动台阵方法和单点谱比法联合的区域速度结构探测方法,设计了多个子群体同步独立进化,定时交换进化成果的伪并行遗传反演策略。针对区域近地表三维速度结构的探测,提出了地脉动联合反演方法。联合反演包含多场地的联合、地脉动台阵方法与单点谱比法的联合。多场地联合反演,假设所有场地相应土层具有相同的剪切波速,考虑了各场地土层结构的共性和生成上的联系,同时减弱了单个场地观测、频散曲线提取和反演等误差的影响;各土层的剪切波速和在每个场地中的厚度通过台阵方法和单点谱比法联合反演获得,充分利用单点谱比的基阶频率信息与覆盖层厚度的良好对应关系,弥补了台阵方法对基岩深度不敏感的弱点,很好地反映了每个场地的自身特殊性。联合反演中利用了已达成共识的基阶卓越周期,比已有研究中采用有争议的瑞利波椭圆率要更为合理。针对联合反演中,反演参数大幅增加,反演效率降低和多解可能性增加等问题,设计了伪并行遗传反演方法,采取多个子群体同步进化,运行参数各不相同,对解空间“探测”和“开采”各有侧重,充分保持了种群的多样性,有效降低了早熟现象的发生。同时,子群体相互之间还定时交换最新进化成果,提高了反演效率。3.建立了一种根据分布稀疏且不均匀的探测数据进行区域场地近地表三维速度结构成像的双调和格林张力样条函数方法和克里金方法相结合的两步空间预测方法,根据地脉动探测数据实现了唐山及周边地区近地表三维速度结构成像。本文发展了两步空间预测方法。首先,固定各土层的平均波速,利用双调和格林函数作为样条函数进行研究区域各土层间界面的空间预测,保证界面的相对光滑性;提出了一种基于地脉动单点谱比法的张力系数确定方法,在充分利用地脉动单点谱比信息的同时避免了张力系数取值的随意性。第二步,利用克里金方法深刻发掘速度结构作为区域化变量的变异本质,充分考虑结构特性,通过高精度的空间外推和预测,很好地模拟了界面间速度结构的空间连续分布格局,最终形成空间二维、三维体精细结构模型。这种两步空间预测方法,充分考虑了场地速度结构自身区域化变量特点和地脉动探测数据相对稀疏和分布不均匀的特点,除用于地脉动探测成果外,还可以方便地吸纳其它各种方法速度结构探测的成果。4.通过参加ESG2006盲比测验、唐山26个联合观测场地的数据分析,与国际领先的地脉动和表面波谱分析方法的比较验证,评价了本文发展的地脉动分析方法的测试精度,通过对结果的分析,得出诸多有益结论和认识。作为一种在实践中发展起来的方法,在实用中检验是至关重要的。通过在唐山26个场地与表面波谱分析方法的国际领先水平比较和参加ESG2006组织的盲比测验,全面地检验了本文整套方法的精度。结果表明,不论人工合成记录还是实际观测记录,不论是速度还是加速度记录,本文方法得到的场地瑞利波频散曲线,均能够为后续反演研究提供可靠的目标曲线;可以根据仪器实际情况自由选择观测加速度或速度。人工合成数据分析中,有两例频散曲线受到了高阶振型影响,在实际观测数据中影响不明显。通过对前人结论的分析,结合唐山26个场地和ESG2006盲比测验结果,提出空间自相关方法的可探测波长范围为Rm in <λ< 4Rmax,为观测台阵的设计提供了指导依据。反演的速度结构可以达到与目前工程勘察中广泛应用的表面波谱分析方法的国际领先水平相同的精度,且在探测深度上有优势。总之,本文从频散曲线提取、反演策略和三维速度结构成像等三方面入手,先分头研究,再整合提高,发展了一套完整的场地近地表三维速度结构的地脉动探测和成像方法。更多还原

【Abstract】 Site condition governed distribution of strong ground motion and earthquake damage significantly. The studies on this issue have been important in the areas of engineering seismology and seismology for decades. Many problems on ground motion are associated with in the lack of site condition information of the observation stations. To detect 3D subsurface Vs structure of engineering site by observing microtremors, must be a powerful tool for development in this subject of Earth Science. In this thesis, the exploration and imaging of the 3D subsurface Vs structure by means of microtremor approaches are studied. The main achievements are summarized as follow.1. An immune-heuristic genetic inverse strategy for deducing Vs structure from the dispersion curves of Rayleigh wave is developed, which combines float coding with roulette selection, arithmetical uniform crossover and mutation into a basic genetic inversion frame, adds the immune heuristic method as local maximum searching scheme and adopts no-repeating forward calculation as speedup scheme, and determines the final structure from the best one of the local optimum by conducting Simulation Annealing algorithm to the result of each individual run. Various combinations of genetic operator and codings are examined by the virtual inversion, as the result it is pointed out that the offline capability is best in dispersion curves inversion by combining float coding with roulette selection. It’s suggested to integrate the above two with arithmetical uniform crossover and mutation into a basic genetic inversion frame.In this inversion strategy, the immune heuristic method and the no-repeating forward calculation method are added to the above basic frame. The immune heuristic method speeds up the evolution by adopting the best individual’s information, enhances the local search in its neighboring space by adding one random number with normal distribution, while giving attention to the search in other areas else by rectifying the distribution deviation. The local search and global search are linked effectively, the disturbance to evolution of the genetic algorithm is minimized. The times of forward caculation are reduced greatly, the calculating cost is saved and the evolution efficiency is increased, from no-repeating forward calculation.According to simulation results for complex soil layered condition, the deficiency in assumption of equal small thickness layers in inversion is pointed out, and the scheme to assume the number of layers and take the Vs and thickness of each layer as parameters to be inversed is proposed. The final result is determined from the best of the local optimum by conducting Simulation Annealing algorithm to the result of each individual run, is proposed. The shortage of averaging the results of all individual runs is avoided. The stability and precision of inversion is increased and the non-uniqueness is reduced.2. An exploration approach for regional Vs structure is developed, by jointly inversion of multiple sites and of microtremors array method with H/V method. A pseudo-Parallel Genetic inversion strategy is designed, in which sub-populations with different running parameters are evolved simultaneously and results in evolution are exchanged timely. A jointly microtremors exploration approach for regional Vs structure is developed. The joint inversion includes the joint of multiple sites and the joint of the microtremors array method and H/V method. The Vs of each soil layer keeps the same at the all sites, to consider the commonness of the soil structure and the depositing connection in the region, and the influence of the errors in observation and in dispersion curves extraction and inversion in individual site can be trailed off. The Vs of each layer in the region and the thicknesses of soil layers in each site are inversed jointly by combination of the microtremor array method and H/V method. By this way, the excellent correlation between the fundamental frequency of microtremors H/V and the thickness of sediment is fully exerted to make up the shortage in array method which is not sensitive with the depth of the buried rock surface, and the speciality of Vs structure at each site is well described. In the inversion procedure, H/V fundamental frequency, which is commonly recognized as an index of soil depth, rather than the ellipticity of Rayleigh wave which remains in dispute, is adopted as the target reasonably.A pseudo-Parallel Genetic algorithm is designed, against the number of parameters to be determined increasing greatly and the inversion efficiency reducing and the possibility of non-unique increasing, in the inversion. In the pseudo-Parallel Genetic algorithm, sub-populations are evolved simultaneously with different running parameters to emphasizing respectively on‘exploration’or‘exploitation’. The diversity of the population is holding while the premature is avoided. The inversion efficiency is improved, by exchanging results in evolution timely.3. A two–step approach for 3D Vs structure imaging from sparsely and non-uniformly distributed data by means of integrating the interpolation method of biharmonic Green function as spline in tension with Kriging method, is established. 3D subsurface Vs structure in Tangshan and surrounding region is imaged by microtremors survey data.A two-step space prediction approach is developed in the dissertation. Firstly the average Vs of each layer are fixed, the interfaces between each two soils layers are predicted by the interpolation method of biharmonic Green function as spline in tension, to keep soil interfaces varying smoothly in space. Furthermore, a method to determine tension coefficient is proposed based on microtremor H/V method, which makes most use of H/V information and avoids the determination of the tension coefficient at will.Secondly, the variance essence of the Vs structure as a regional variable is deeply excavated by Kriging method and its structural characteristic is taken into account. The spatial successive distribution of the Vs structure between soil interfaces is simulated by high precision space extrapolation and prediction, and finally the detailed 2D or 3D model is constructed.The two-step space prediction approach describes the characteristic of Vs structure as a regional variable from the sparsely and non-uniformly distributed microtremors data. Obviously, it also can easily make good use of the other exploration method.4. The microtremor method proposed in this dissertation is validated by comparing with the international advanced microtremor methods and SASW methods via ESG2006 Noise Blind Test and results of microtremors observations at 26 sites in Tangshan region. Some understandings and conclusions crucial for guiding microtremors array observation are also drawn.As an exploration method developed in practice, it is necessary to be validated in test. The method presented in the paper is tested with results from the international advanced approach of SASW at 26 sites in Tangshan region and ESG2006 Noise Blind Test. The results show that the method obtains reliable results either from artificial records or in-situ observation records in acceleration or velocity . Either of them can be chosen in observation contingent on the capability of instruments. Although high mode of dispersion curves are observed in 2 cases in artificial records analysis, but in microtremors records observed in-situ it is not notable. Based on analysis of the former researchers’conclusion and the above results, the reliable wave length range of SPAC is proposed as Rm in <λ< 4Rmaxto instruct microtremor array design. The Vs structure inverted may reach deeper with the same precision as the international advanced approach of SASW, well applied in geotechnical exploration. In conclusions, an integrated microtremors method to explore and image 3D subsurface Vs structure is developed in this dissertation. The dispersion curve extraction, inversion strategy and 3D structure imaging are studied separately, then integrated and improved.更多还原