【作者】 廖力；

【导师】 周建中；

【作者基本信息】 华中科技大学 ， 水利水电工程， 2013， 博士

【摘要】 洪水灾害是自然界一种矛盾集合体，它一方面具有极强破坏力，对人类社会和自然环境造成巨大损害；另一方面又是维持自然生态平衡的必要环节。国内外治水的经验教训证明，单纯的―防御洪水‖策略是没有出路的，并且随着流域经济建设规模的不断加大和多种因素导致的洪水规模的不断上升趋势，来自于防洪风险的威胁正变得越来越严重。在遭受了惨痛的教训后，学者们进行了反思，提出了洪水灾害风险管理的理念洪水灾害成灾机制的复杂性、致灾过程中的随机性以及承灾体的差异性决定了洪灾等级划分的难度巨大。但是近年来随着特大灾害频发且危害巨大，出于灾害预报和御灾减损的需要，灾害等级评估工作作为洪水灾害风险管理的重要环节越来越受到重视。本文在研究多级模糊综合评判理论与多维指标定量类别划分原理的基础上，以洪水灾害等级评估为研究对象，对如何实现标准缺失的复杂随机洪水样本集准确分类，建立具有普适性指标、跨时空性和高认可度的洪水灾害等级评估标准体系，明确分级标准存在条件下大信息量、高精确度的洪水灾情综合评估等科学问题进行了研究与探讨，给出了解决方案并进行了有效性及可靠性验证。相关研究成果已应用于973项目的工程应用示范中。本文的主要研究内容和创新包括：（1）针对模糊聚类迭代方法对于样本分布的依赖性问题，将核函数方法应用于模糊聚类模型中，通过核映射将样本空间映射到高维特征空间，并在高维特征空间推求线性回归方程，使处理后的样本更适合于聚类运算，有效提高聚类效果和聚类准确率，且其计算复杂度不会随着特征空间维数的增加而有明显变化。（2）针对样本集中出现超大样本时模糊聚类模型中聚类中心会出现极大偏移导致分类效果急剧转劣，提出了以样本值对于平均值的加权相对距离及切比雪夫不等式为判据的超大样本辨识方法，并给出了存在超大样本的洪灾样本集的准确评估方案。（3）针对目前具有强普适性、高认可度的洪灾评估标准缺失的问题，深入研究了模糊聚类迭代方法的理论与流程，利用聚类中心矩阵及模糊分类的欧式权距离判别依据，开发了模糊聚类迭代的标准制定功能，并在此基础上进行了洪水灾害等级评估标准制定的建模工作，标准制定过程中进行了指标及样本值的二次标准化，保证了所定标准的普适性。模型经过实例验证是可行且有效的。（4）针对现有投影寻踪模型或参数设置主观性强或对于训练样本及经验值过于依赖的问题，提出了对于投影寻踪聚类效果有着更好解释的投影指标函数，构建了全新的投影寻踪聚类模型，极大提高了聚类客观性与评估效果的同时，大大降低了运算量，结合多项式函数可以得到待评估样本的精确连续性等级值（5）针对传统优化算法对于洪灾评估模型中存在的优化问题寻优效果不佳的问题，提出了与文化算法相结合的自适应混沌差分进化算法，利用文化算法的学习与进化能力，提高差分进化算法搜索效率，同时采用自适应的变异因子和交叉因子，改善了算法的收敛能力。更多还原

【Abstract】 Flood disaster is a kind of natural phenomenon. The paradox is that on the one hand ithas a strong destructive power, cause great damage to human society and naturalenvironment, and on the other hand it is essential for maintaining natural ecologicalbalance. The experience and lessons of water conservancy in the domestic and overseasshow that the kind of pure "flood preventing" strategy is no way out. And, with theincreasing of the economic construction scale near the basin and the rising trend of thescale of flood caused by many complicated factors, the threat of flood is becoming moreand more serious. After suffered a painful lessons, the scholars put forward the concept offlood disaster risk managementThe complexity of flood mechanism, the randomness in the process and the differencesof hazard-bearing body determines the immense difficulty of floods classfication. But inrecent years with the frequent occuring of heavy disasters and its great harm, as animportant part of flood disaster risk management, disaster grade evaluation is paid moreand more attention for disaster forecasting and royal disaster loss. Based on the researchon the theory of multistage fuzzy comprehensive evaluation and the principle ofquantitative multi-dimensional multi-index classification, the dissertation take the flooddisaster level assessment as the research object, study and discuss the questions that howto set complex random samples‘accurate classification under the condition of standardslackness, how to set up a flood disaster level assessment criteria system with universalflood index, the space and time acrossing attribute, higher recognition, and how to assessthe flood samples with large amount of information and high accuracy under a clearclassification standard. Some solutions are given and their validity and reliability areverified. Related research achievement has been applied in engineering application. Thisarticle‘s main research content and innovation can be described as follows:(1) According to the dependence on sample distribution of the fuzzy clustering iterationmethod for of, kernel function method was applied in fuzzy clustering model, through thekernel mapping, the sample space is mapped to a high-dimensional feature space, andlooking for linear regression equation in high dimensional feature space, making thetreated sample more suitable for clustering algorithm, effectively improving the effect andaccuracy of clustering. And its computation complexity is not obvious increased with theincrease of the feature space dimension.(2) In order to solve the problem that classification effect will rapid decrease with theoccuring of a super big sample which cause large deviation of the fuzzy cluster centers,this article put forward the identification method of the super big samples by the weightedaverage relative distance and chebyshev inequality.(3) Aiming at the missing of the flood disaster evaluation standards which have stronguniversality and high recognition, this article deeply studied the theory of fuzzy clustering iteration method and process, develop the standards formulating function of fuzzyclustering iteration model based on the clustering center matrix and fuzzy classification ofeuropean-style distance discriminant and build a flood disaster level assessment standardsformulating model. In the modeling process, indicators and the sample values carried ontwice standardizing to ensure the universality of the standards. Model is verified to befeasible and effective through example.(4) In view of that the strong subjectivity in parameter setting and high dependence oftraining samples and the experience value in existing projection pursuit, this article putforward a new projection index function with a better explanation for projection pursuitclustering effect and build a new projection pursuit clustering model, which greatlyimproves the clustering objectivity and evaluation effect and reduces the computationalcomplexity. The model can get a precision continuity level value of a sample bycombining with polynomial function.(5) In consideration of that the traditional optimization algorithms perform poorly in theoptimization problem in the flood classification model, an adaptive chaos differentialevolution algorithm with the combination of cultural algorithm is put forward. With thelearning and evolution ability of the cultural algorithm, the new algorithm improves theefficiency. Meanwhile, by using adaptive mutation factor and crossover factor, theconvergence ability of the algorithm is improved.更多还原