基于模糊投影寻踪聚类的洪灾评估模型

针对洪水灾害样本集的复杂性、随机性以及差异性,本文将模糊聚类迭代理论与投影寻踪技术进行互补融合,构建了模糊投影寻踪聚类模型.该模型采用投影值标准差和投影值欧氏距离平方和来构造投影指标函数,避免了传统投影寻踪模型由于经验性选取密度窗宽导致过于主观的问题;将高维数据低维化后进行模糊聚类运算,再通过对最优投影方向的寻优进行投影寻踪聚类运算,降低了模糊聚类迭代的运算量,实现了两种模型的双重迭代聚类.误差分析及聚类有效性评价表明,互补融合后模糊聚类与投影寻踪聚类的双重迭代聚类使得聚类精度和效果也得到了较大提高;此外,除优化算法初始化参数外,模型无需预设其它参数,也不依赖于随机训练样本,可客观依据历史样本集内在规律来进行洪灾聚类评估.实例研究和方法比较表明,提出的模型计算简洁,且能有效处理洪灾评估中的随机、模糊等主客观不确定性,能够为洪灾风险管理提供科学的决策支撑.

A fuzzy projection pursuit clustering algorithm based flood disaster assessment model

In view of the complexity, randomness and diversity of flood samples, this paper constructed the fuzzy projection pursuit clustering (FPPC) model which was combined with the fuzzy clustering iterative model and the projection pursuit technique. The model, for the first time, adopted a new projection index function formed by the standard deviation of projection value and the quadratic sum of Euclidean distance of projection value, which has a better validity to the clustering target and avoid the empirical selection of density window width. The FPPC model clusters the samples after reducing their dimension by projection technology, significantly not only reduces the fuzzy clustering iteration computation, but also improves the clustering accuracy and effect. In addition to the initial parameter of optimization algorithm, the model does not need to set any other parameters. Based on the assessment on the historical flood sample set and the comparison with other methods, this paper has verified the accuracy, objectivity, and validity of the FPPC model. The research results are expected to shed a light on making effective decisions for flood risk management.