【作者】 孙玮；

【导师】 王树良；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2013， 博士

【摘要】 地质灾害的发生不仅威胁着人类的生命,而且还对人类的生存环境和资源等造成巨大的破坏,其中最频繁发生的滑坡所造成的影响也是极其恶劣的。研究发现,影响滑坡发生的因素有很多且都具有不确定性和随机性,各自变化很难寻找统一的规律,使得归纳不同类型不同区域的滑坡的发育演化过程以及对斜坡的稳定性进行评价变得相当困难。同时,稳定性评价指标是高维非正态数据和非线性数据的组合,而常规的滑坡评价方法无法适应在无条件的情况下对高维非正态、非线性数据的分析,难以得到应用价值较高的理想效果。至今,研究滑坡预防预报的工作者们的共同目标就是分析有用的滑坡数据,提取其中的潜在规律。在研究滑坡的物理形成机理的同时,发现滑坡发育形成的过程相当复杂,而且难以精确地表达的,滑坡是先由什么引发再恶化,最终致伤及生命、损毁房屋、公共设施等。滑坡数据挖掘已经成为滑坡研究工作者们的工作重点和方向,对于那些由于监测系统越来越完善和良好的通信设备的安置获取的实时监测数据,研究人员表现出了浓厚的兴趣。不同的数据挖掘算法分别应用于这些数据,得到了很多有价值的信息,但是算法的复杂性和计算量大的特点,对于海量的不同数据源所获取的数据来说,很难被人们理解和更好地运用。为了在一定程度上缓解这些问题,研究员们将数据挖掘技术和可视化技术有机融合在一起,实现两种技术的相辅相成,这种融合使得数据挖掘技术的应用得到更加直观和形象的展示,它将人们的认知能力、创造力、不同领域的知识与数据挖掘的过程有机地结合,并充分发挥两者的优势,不仅体现了用户的核心作用,而且也提高了数据挖掘的效率和精度。针对影响滑坡稳定性因素的不确定性和随机变化的特性,同时由于滑坡监测的环境恶劣且安置监测装置的不便造成监测数据的不完整性和其间的冗余现象,本文制订了对滑坡数据的处理方案,对滑坡稳定性影响因素中的非定量数据进行了量化,然后对那些不必要的影响因素进行约简和删除,建立更具有实效性的滑坡稳定性影响因素指标体系。本文将可视化数据挖掘技术应用于滑坡形成机理研究,进而建立可视化滑坡预警分析模型,针对多种数据分析算法和预测模型在不同的滑坡稳定性评价中的缺陷,讨论了基于平行坐标可视化方法建立滑坡预报预警分析模型的思想,并把这种方法应用于滑坡稳定性评价中。利用各种滑坡监测数据和滑坡点的历史滑坡发生数据联合分析,在平行坐标轴上显示和分析这些数据,并且利用平行坐标轴中的刷技术、交换坐标轴、上卷下钻、数据抽象、坐标轴放缩等分析方法来找出相关联的数据项,通过直接聚类分析和交互式聚类分析来发掘数据间的隐含关系和连带因素,将平行坐标可视化与聚类分析有机结合,完成了滑坡预报预警模型的设计。同时,针对平行坐标可视化方法的数学基础以及映射规则,讨论了基于平行坐标可视化的滑坡预报预警评价等级标准的制定。最后,为了更好地说明此种方法适用于绝大多数滑坡的个性特点,体现出针对不同滑坡个体预警建模的通用性,选用了一些以往的滑坡监测数据和历史灾害点和隐患点排查数据来进行分析。实验说明,平行坐标可视化方法在灾害数据的分析中得到了充分地发挥,不仅对建立该滑坡点的监测体系和评价指标体系有重要的指导作用,还建立了容易操作和易被用户理解的预警模型,为防灾减灾工作做出了贡献。综上所述,本文将理论简单且容易接受的平行坐标可视化方法应用于滑坡监测预警分析建模中,不仅实现了所有滑坡监测数据的实时可视化,也实现了空间-时间序列分析同时进行的设想。实验表明,此方法适合于不同地域的不同类型的滑坡分析和建模,不仅简化了滑坡分析的数学建模过程,也节省了大量算法的复杂运算时间,使得分析建模的过程在完全可视化和更容易理解的基础上完成,为相关部门及时作出防灾治灾决策奠定了坚实基础。此方法在滑坡预报预警方面的应用,不仅对滑坡的个体特点,而且对外界条件的影响做了深入分析,由于人为干预的影响使得分析建模的速度和预报预警的精确度和效率都有一定程度的改善,为挽救人类的生命和财产做出了应有的贡献。更多还原

【Abstract】 Geological disasters not only threaten human life,but also on the human living environment and resources cause enormous damage, landslides occurred most frequently in which the impact is extremely bad. Studies found that there many factors affecting landslide and have a lot of uncertainty, every factor keep change and difficult to find a uniform law. Therefore it is difficult to make the induction for the law as well as the occurrence of landslides in an area of landslide stability evaluation. Meanwhile, stability evaluation depend on high dimensional, non-normal and nonlinear data,while the conventional evaluation methods for these data is not strong, it is difficult to obtain the desired results. For landslide data analysis and extraction of useful potential law, is researchers’ aim. During the formation mechanism of landslide physical process, the landslide process is quite complex and difficult to accurately represent what is to be preceded by landslide injuries caused further deterioration resulting in life, damaged housing and public facilities.Currently, the landslide data mining has become a researcher’s work focus and direction. As for those monitoring systems become more perfect and good placement of communications equipment for real-time monitoring data, the researchers demonstrated a strong interest. Different data mining algorithms are applied to these data, get a lot of valuable information, but the algorithm complexity and computationally intensive characteristics of mass data obtained from different data sources it is difficult to be understood and more good use of. Data mining and visualization technology have been made to solve these problems, organically together to achieve the two technologies complement each other, this integration makes data mining technology to be more intuitive and graphically display, it organically integrates people’s cognitive abilities, creativity, different fields of knowledge and data mining process, and give full play to the advantages of both, not only reflects the central role of the user, but also improves the efficiency and accuracy of data mining.For the stability of the landslide affected by many complex factors, and the data collected have the features of not-comprehensive or redundant, in order to establish a stable indicator system, this paper first quantified the factor that affecting the stability of the landslide, then simplified and deleted the properties of those unwanted items. For a variety of deficiencies of data analysis algorithms and predictive models in different landslide stability evaluation, this paper discussed the thinking of establishing landslide early warning and forecast analysis model based on the basic principles of visual data mining and the features of parallel coordinates visual data mining, and applied this method in landslide stability evaluation. Then did combined analysis using a variety of landslide monitoring data and historical data, and using the parallel coordinate axes brush technology, switching axes, up-scroll and drill-down, abstract and other analytical methods to identify the associated data items, through using direct cluster analysis and interactive clustering to discover the implicit relationships and associated factors between two parallel coordinate axes. Ultimately, we combined data mining method of parallel coordinates visualization with clustering analysis, completed the design for landslide forecasting and early warning modeling. At the same time, in view of mathematical foundation and mapping rules of parallel coordinates visualization method, we discusse and develope the evaluation grade standard of landslide forecast and early warning based on parallel coordinates visualization.In order to illustrate that this method is applicable to a variety of regional landslide, reflect the general characteristics of landslide early warning model, we selected some previous landslide monitoring data, historical disaster and hidden danger investigation data for analysis. Experiments show that the parallel coordinates visualization method give get full play to analyze these data. It is not only has the important instruction function to the establishment of the landslide monitoring system and evaluation index system, it also establishes the early warning model, which is easily to be operated and understood by non-professional users, contributed to disaster prevention and mitigation.This article applied simple and easy to accept parallel coordinate visualization method to landslide monitoring and early warning analysis modeling, not only achieved all real-time visualization of landslide monitoring data, but also achieved a vision of space-time series analysis simultaneous. The results of experiments shows that this method is suitable for different regions of different types of landslide analysis and modeling, not only simplifies the mathematical modeling process, saves much computing time for severing people, but also making the analysis of the modeling process in complete visualization and easier to understand, then laid a solid foundation for relevant departments to make timely decisions to prevent disaster. This method can improve the speed of landslide modeling, upgrade the accuracy and efficiency of forecasting and warning, and then save human lives and property.更多还原