不确定DM-chameleon聚类算法在滑坡危险性预测的研究及应用

发布时间：2018-03-14 23:20

本文选题：滑坡　切入点：危险性预测　出处：《江西理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：滑坡灾害是我国乃至世界范围内发生次数最多的地质灾害之一,它不仅对财产、环境和资源等产生破坏性,而且会给人类生命安全带来严重威胁。滑坡灾害危险性预测研究是一个复杂的多源信息综合分析过程,其中所含信息具有突变性、非线性、随机性和不确定性等特点。这些特点给滑坡危险性预测带来相应的困难。由于滑坡的频繁发生,所造成的危害形势也日趋严峻,因此寻求一种科学有效的方法来提高滑坡预测的准确性是一个很有意义的研究课题。数据挖掘的相关理论方法有较强处理非线性关系的能力,例如通过聚类方法可将高度相似的数据对象归为在同一类中,高度相异的数据对象分在不同类中,从而依据已知规则对未知事物进行预测。因此本文采用数据挖掘中Chameleon聚类算法和滑坡信息的有关特点相结合,构建滑坡危险性预测模型,设计聚类子集危险性等级划分方法,进而对实例研究区滑坡危险性等级进行预测划分。研究发现滑坡灾害是由坡高、坡型、和降雨等多种因素共同作用所引发的,其中与滑坡发生机制有着密切关系的降雨量取值在一个不确定区间内,具有不确定属性,致使传统Chameleon聚类算法很难准确对其进行定量刻画及有效处理。而且传统的Chameleon聚类算法也存在构建k-最近邻图kG时k值的确定与相似度函数阈值的选取都需要人工进行和处理大规模数据集的局限性等问题。为了解决上述所存在的相关问题,本文基于前人所研究的M-chameleon聚类算法基础上,提出一种新的两阶段聚类整合算法(DM-chameleon)适用于处理较大规模数据集;引入不确定数据模型,有效的利用不确定属性的特征刻画降雨量;并对聚类技术中表征相似性的欧氏距离进行拓广,使其适用于不确定数据之间相似性的计算;最后根据上述理论提出一种不确定DM-chameleon聚类算法并把其应用在滑坡危险性预测模型中,以延安宝塔区为实例进行验证。首先在已知数据集上实验得到DM-chameleon算法比M-chameleon算法获得了较好的聚类效果,并且聚类速度有了明显提高,其次对比实例研究结果表明不确定DM-chameleon聚类模型取得了较高的预测精度,进而验证了不确定DM-chameleon聚类算法应用在滑坡危险性预测中的可行性。
[Abstract]:Landslide disaster is one of the most frequent geological disasters in China and the world. It not only destroys property, environment and resources. The study of landslide hazard prediction is a complex process of comprehensive analysis of multi-source information, in which the information contained in it is abrupt and nonlinear. The characteristics of randomness and uncertainty bring the corresponding difficulties to the prediction of landslide risk. Because of the frequent occurrence of landslide, the harmful situation caused by landslide is becoming more and more serious. Therefore, to find a scientific and effective method to improve the accuracy of landslide prediction is a meaningful research topic. For example, by clustering, highly similar data objects can be classified into the same class, and highly different data objects can be divided into different classes. So this paper combines the Chameleon clustering algorithm in data mining with the characteristics of landslide information, constructs the landslide hazard prediction model, and designs the method of classifying the risk levels of clustering subsets. The landslide hazard is caused by various factors, such as slope height, slope type, rainfall and so on. The rainfall which is closely related to the occurrence mechanism of landslide is in an uncertain range and has uncertain properties. It is very difficult for traditional Chameleon clustering algorithm to accurately characterize and deal with it effectively. Moreover, the traditional Chameleon clustering algorithm also needs to determine the k value and select the threshold of similarity function when constructing k-nearest neighbor graph KG. Problems such as the limitations of large data sets are carried out and dealt with. In order to solve the related problems mentioned above, Based on the M-chameleon clustering algorithm, a new two-stage clustering integration algorithm (DM-chameleon) is proposed to deal with large scale data sets, and an uncertain data model is introduced to characterize rainfall effectively. The Euclidean distance which represents the similarity in clustering technology is extended to make it applicable to the calculation of similarity between uncertain data. Finally, according to the above theory, an uncertain DM-chameleon clustering algorithm is proposed and applied to the landslide hazard prediction model. Taking Baota area of Yan'an as an example, the experimental results show that DM-chameleon algorithm has better clustering effect than M-chameleon algorithm, and the clustering speed is improved obviously. Secondly, the results of the case study show that the uncertain DM-chameleon clustering model has achieved high prediction accuracy, and the feasibility of applying the uncertain DM-chameleon clustering algorithm to landslide hazard prediction is verified.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P642.22;TP311.13

【参考文献】