【作者】 陈晓利；

【导师】 叶洪；

【作者基本信息】 中国地震局地质研究所 ， 构造地质学， 2007， 博士

【摘要】 地震滑坡是一种常见的地震次生灾害形式，以其巨大的致灾力引起人们的广泛关注，在山岳地区，所造成的损失有时甚至远远超过地震本身。目前，世界上很多国家都在进行地震滑坡潜在危险性研究，其意义在于潜在地震滑坡危险区的确定，使我们在进行基础建设和区域规划时，有依据选择合适的场地，避开危险地段，或者采取必要的防范措施，从而达到减少经济损失，保护人民生命财产安全的目的。我国是一个多山、多丘陵的国家，据统计，山地和丘陵面积约占国土面积的70％，这就从客观上决定了我国有大量的自然边坡。我国西南地区区域地质背景复杂，是中国大陆内强震活动频度最高的地区。据两千多年来的历史资料记载，西南地区曾发生过很多强烈的地震，引发的滑坡、崩塌问题特别严重。我国正处于开发西部的征程中，由于经济的发展，对土地的需求在不断地增加，对土地的合理利用提出了迫切的要求；同时，西南地区许多重大水利水电工程正在建设中或处于论证阶段，而这些重大工程的选址又常常在高山峡谷中，对边坡的稳定性研究显得更为重要。因此，对地震滑坡、崩塌问题的研究具有不仅有重要的理论意义，而且具有重要的实践意义。地震滑坡作为滑坡的一种类型，有其自身的特点，本文通过对大量有地震滑坡纪录的震例分析中，系统地总结了地震滑坡的分布特征、形成条件和相关影响因素。在前人研究成果的基础上，研究了地质构造背景、岩石结构、岩性特点、地形地貌、水文条件等对地震滑坡的影响，同时对地震滑坡与地震动参数之间的关系进行了论述。论文对现有的斜坡稳定性评价方法进行了回顾，并针对目前区域性地震滑坡稳定性评价所用方法中存在的所需工程地质参数太多、数据获取困难、赋值主观性较强等缺陷，本文进行了一系列的改进。在本文的研究过程中，地理信息技术(GIS)的应用对地震滑坡研究起了极大的推动作用。一方面，地震滑坡是由地震触发的，在分布上具有量多面广的特点，这种区域特点的问题适宜于GIS对空间数据管理的特点；另一方面，地震滑坡的影响因素众多，各个因素之间互相影响、互相牵掣，传统的数据分析很难把这些不同来源、不同性质的数据集中分析，而GIS空间数据库功能可以把各种影响因素搁置于统一的地理平台进行讨论，在数据相关性分析上具有不可比拟的优势。GIS把地质、地貌、岩性、构造、植被、降水等与地震滑坡相关的环境资料一起储存在空间数据库中，并应用GIS的空间分析功能对这些数据进行分析研究。GIS的引入，使得对滑坡的研究不再是孤立地研究单一因素与滑坡的关系，而是把滑坡事件与周围的地质、地貌环境等资料综合起来进行分析。地震滑坡形成机制复杂，涉及因素众多，它在空间上不是完全随机分布的，换言之，地震滑坡的影响因素和它的分布规律之间存在着相关性。为了表达这些特征因素与地震滑坡发生的关系，本文利用径向基概率神经网络自学习、自适应的特性，通过对样本训练、检测，最终得到一个稳定可靠的模式识别网络，从而通过该模式对研究区域的地震滑坡进行识别。应用神经网络研究地震滑坡危险性预测是有其理论基础的。从工程地质学构造类比的角度讲，对潜在地震滑坡危险性进行判断，实质上是一种模式识别问题。神经网络方法的运用使得对这一事件的认识更客观、更接近实际。本文在对地震滑坡数据进行空间分析的基础上，结合前人的研究成果，采用易于获取的信息资料，包括水系、断裂、岩性、坡度、地震烈度等5项指标作为地震滑坡危险性研究的神经网络的输入指标。在对地震滑坡危险性评价的工作中，影响因素权重大小以及划分危险性级别的各指标界限的确定问题上的模糊性，决定了该工作的复杂性。传统的确定各个影响因素重要性的方法是根据专家的经验，避免不了主观因素的影响。层次分析法则提供了一种确定权重的较好方法。它通过两两对比的方式，确定各个因素的相对重要性，基于统一的标准建立判断矩阵进行综合判断，最终可得出各因素按其重要程度的排序。在危险性界别的划分上，通过建立单因素指标评价矩阵，与层次分析法确定的影响因素的权重向量进行模糊合成，最终根据最大隶属度确定危险性所属级别。这一方法的实施，解决了危险程度划分界限不能明确表达的难题。为了对上述理论的实际应用性进行检验，本文以发生在我国西南地区的3个强烈地震(1973年炉霍地震，M=7.9；1974年昭通地震，M=7.1；1996年丽江地震，M=7.0)为例，应用神经网络、层次分析模糊数学对震区的地震滑坡危险性进行了研究。(1)在研究各个地震震区滑坡分布规律的基础上，对每一震区选出各自的训练样本建立相应的网络模型，然后对整个区域进行识别。各个震区的识别结果表明了神经网络对地震滑坡单元具有良好的识别能力；(2)在单独对各个震区的地震滑坡进行识别的基础上，合并各个震区的训练样本，用统一的网络模型对研究区域的地震滑坡进行研究。结果表明，统一的网络模型在各个区域均取得较好的识别效果；(3)对地震滑坡的5个主要影响因素水系、断裂、岩性、坡度、地震烈度等，根据本区地震滑坡分布规律与它们之间的关系，采用层次分析的方法，确定了这5个因素之间的相互重要性并建立单因素的评价标准。本文中这5个影响因素的权重向量W=(0.0491 0.1379 0.3393 01850 0.2855)，进而通过模糊合成形成判断集B=WOA，即可对每个单元进行等级判断。从最终结果中可以看到三起地震中实际滑坡发生的位置大部分均在本文划分的高度危险区中。从上述两种方法的理论基础并结合实际应用中取得的成果来看，可得到如下的认识：本文选用的地震滑坡影响因素作为神经网络的输入特征指标具有科学性和实用性，根据3起地震提取的样本训练而成的神经网络模型，可用以对有相似地质构造背景的地区进行地震滑坡危险性预测。不同构造区域中，相同影响因素的权重可能不同，也就是说，在一个区域中具有较大影响的因素在另一个区域中的作用可能不是很大。这从一个方面反映了地震滑坡的复杂性。更多还原

【Abstract】 As a kind of secondary disasters caused by strong earthquakes, the earthquake-induced landslide has drawn much attention in the world because of severe hazards it causes. In a mountainous region, sometimes a great loss of lives and properties caused by landslide even exceeds the losses caused by the earthquake itself. These large landslides usually cannot be prevented by current mitigating measures, whereas the only possible preventive measures are early warning and evacuation of vulnerable communities. In order to reduce the damage, researches on the potential earthquake-induced landslide zoning are conducting in many countries at present. The purpose of the zoning map is to provide a tool for regional planning. Based on it, the government can avoid the dangerous areas and select suitable sites for constructions, thereby protect peoples’ lives and properties.In southwest China, due to its complex geological and geographical conditions, many strong earthquakes have occurred, inducing lots of landslide, therefore, earthquake-induced landslide remains one of the most serious seismic hazards there. So, the study of this issue is very important in theory as well as practice especially when our government is implementing the grand strategy to develop western China.Supported by hundreds of bibliographies, this thesis systemically details the earthquake-induced landslide’s distribution characteristics, formation conditions and related influencing factors. Based on previous research results, this work summarizes relationships between landslides and influencing factors, such as geological background, rock mass structure, topography as well as hydrogeology condition. Also, this thesis gives a brief glance to the methods used in the slope stability study and puts forward some improvements to the old methods. In this work, Geographic Information System(GIS) technology has been used as a power tool in the research. The question concerned is the spatial features that meet the GIS capacity very well. Spatial database has the ability of controlling all the data from diverse sources, and the factors influencing earthquake-induced landslide which are taken into account have different combinations as known. So, in this thesis, all the factors have been studied together on a uniform platform supported by GIS instead of studying single factors independently.The studies show that earthquake-induced landslide is not distributed at random, rather it has its regularity. It means the corresponding relationship between the earthquake- induced landslide and the geological factors, though it is a non-linear relationship. This work uses neural network named Radial Basis Probabilistic Neural Network(RBPNN) to study this non-linear relationship through the training of landslide samples. In terms of geo-technological structure analogy, to determine the potential landslide places is essentially a pattern recognition question, because the areas where earthquake-induced landslides occur have similar geology conditions. Through repeated sample training, neural network could obtain the model of relationship, and then the model can be used to simulate the potential area that would be influenced by earthquake-induced landslide. The non-linear relationship obtained through neural network training would be more objective than others because the whole process is fulfilled automaticly. In this work, rivers, faults, rock, slope angle and seismic intensity are taken as the neutral network input indexes into account of the existing knowledge of the landslide.It is known that there are many factors influencing earthquake-induced landslide, and different factors have different actions to the same thing. Due to both a great variation in landslide characteristics and our insufficient understanding on mechanisms of landslide, traditional methods in deciding factors’ weights mainly depend on the experts’ experiences, so the results would be subjective to some degree. Analytic Hierarchy Process(AHP) is a good technical approach for converting subjective assessment into a set of weight. It has proven to be very useful in assisting selection from a finite set of alternatives as well as in ranking things. Through comparing factors two by two, pairwise comparison matrix is built, after solving the matrix, weights of relative importance for different factors are obtained. Despite of the complexity of the factor ranking, there is another difficult problem needing to be solved. It is difficult to give a clear boundary for different earthquake-induced landslide hazard grades when we need to tell where the area is more danger than another. Fuzzy Mathematics is good helpful to solve this question. Building a single factor influence matrix according to the grades needed (in this work, the risk is divided into 3 grades: high hazard area, middle hazard area and no hazard area), synthesis it with the weight matrix obtained by AHP, and then a degree of membership matrix is obtained. At last, the hazard grade is decided by the maximal membership degree. The application of AHP and Fussy mathematics could be a useful attempt in the hazard zoning work.In this thesis, three examples are taken and used as target areas to apply the method presented above. They are the Luhuo Earthquake(1973, M=7.9), Lijiang Earthquake(1996, M=7.0) and Zhaotong Earthquake(1974, M=7.1). All of the three earthquakes occurred in southwest China and triggered lots of landslides. Specific steps of analysis are as follows:(1) After the study of every strong earthquake triggered landslides’ distribution, selecting training samples and constructing neural network for every example earthquake, then use the neutral network model to simulate the potential area influenced by earthquake-induced landslide. The results show the good capacity of neutral network in landslide area recognition.(2) Based on the neutral network identifying in different areas, this work takes all the training samples together to obtain a uniform model. The results also show the good capacity of neutral network in landslide area recognition.(3) Based on the study of relationships between rivers, faults, rock, slope angle, seismic intensity and distribution of earthquake-induced landslide, this work decides weights of relative importance for the 5 factors using AHP technique and builds single factor influence matrix(A) at the same time. The weights matrix(W) of the 5 factors is: W= (0.04910.1379 0.3393 01850 0.2855) . Further more, fussy synthesis of 2 matrix is made: B -WOA, where B is a matrix of the final evaluation. This work classes the hazard areas into 3 grades: high hazard area, moderate hazard area and no hazard area. From the zoning results, it’s clear that almost the entire area influenced by earthquake-induced landslide has been enclosed in the high hazard range. In summary, this thesis selects 5 factors out of all influencing earthquake-induced landslide to study its distribution. The results indicate that the 5 factors are scientific and of utility in the neutral network input index and AHP fuzzy mathematics. The neutral network model drawn from the Luhuo earthquake, Lijiang earthquake and Zhaotong earthquake can be used as a tool to identify the dangerous areas which have similar geological conditions and support the land planning. It’s also worth attention that in different geological settings, the same factor may have different actions and different weights.更多还原