【作者】 刘长吉；

【导师】 陈建生；

【作者基本信息】 河海大学 ， 岩土工程， 2007， 硕士

【摘要】 库坝区渗漏破坏的机理始终是工程研究的重要课题之一，本文回顾了国内外多起库坝垮塌事故，其中因渗漏引起的渗透破坏最为常见。本文正是针对这一常见的工程问题进行了探索性研究。对于存在渗漏的库坝区而言，其地下水成分是很复杂的，这种复杂性体现在补给来源多样和受周边环境影响难以消除。本文从分析包括库水、尾水以及场区钻孔中地下水特征在内的水样信息研究渗漏水来源的思路出发，基于模式识别的思想建立相关模型，实现水样类别的定量化划分，并将研究成果应用于龙羊峡坝区地下水渗流场分析。本文针对定量计算方面进行了系统和深入的研究，主要内容如下： (1) 在模式识别思想的基础上将参数无量纲化，从模糊数学的基本理论出发，建立可以为水样定量化分类的混合比模型，并分别从从理论研究和工程实际应用的角度给出了库坝区渗漏相关的水样类别界定标准。根据该模型，可以将常见的水样分为七类，即：理想边坡水、边坡水、边坡水占优的混合水、理想混合水、库水占优的混合水、库水、理想库水。 (2) 针对用于模式识别研究的样本和样本基元的数量、类别等对结果的影响以及它们对于研究的适用性，分别给出样本和样本基元的筛选原则及方法，如模糊聚类方法、统计学方法等。 (3) 基于Talor公式和BP网络的基本算法建立了多因素增量模型，并给出形式简洁、物理意义清晰的解析解，该模型可以通过结合上述混合比模型的水样分类依据完成模式识别。更多还原

【Abstract】 After reviewing some Chinese and foreign accidents in dams and reservoirs region, it is discovered that a great number of them are caused by leakage, which means that the seepage deformation is one of the main reasons that can cause dikes failure, so the study on mechanism of leakage failures has great significance for dike safety, which is also the study center in this paper. Based on the pattern recognition techniques and artificial neural network, a model is established, by which the water samples could be classified as reservoir water, ground water of the well, et al based on the characteristic of water samples. At last, Aiming to test the validity of model, groundwater seepage field of Longyangxia dam was analysised through such model. The major contents are as follows:(1)After to processing non dimensional quantities of original data based on elements of pattern recognition, a mixing ratio model is established based on basic theory of fuzzy mathematics, which can classified the water samples quantitatively. Also the model give the definition criterion of difference water. So the water samples could be classified as pure slope water, slope water local water, local water, main local water mixing water with most slope water, mixing water, mixing water with most reservoir, reservoir water, pure reservoir water based on such criterion.(2)Aiming at making clear the effect degree of result from samples, numbers and types of sample indexes in pattern recognition study field, some filtering principles and methods are put forward, such as fussy clustering method, statistics method, etc.(3)Basing on the basic arithmetic of Talor equation and BP network model, a multi-factor increment model is established. Then, the analytic solution with simple form but clear physic meaning is given. This model may achieve the pattern recognition combining with mixing ratio model.更多还原