【作者】 徐略辉；

【导师】 邵世煌；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 在人工智能、机器学习、模式识别和数据挖掘等领域的研究与应用中存在着不确定性。例如随机性、模糊性,还有不完全性、不一致性等。这些不确定性对于数据分析处理,尤其是在与决策分类相关的属性约简中必须加以考虑,否则可能得不到满意的约简目标。本文研究一种基于模糊粗集等价聚类的不确定性属性约简及其在服装搭配上的应用。主要的研究内容与创新点有:给出了一种模糊粗糙集等价类聚类划分方法;使用了模糊C—均值方法对模糊粗糙集的论域重新进行相关聚类划分,将分类结果作为等价类;得到的等价类的平均基数比较小,能够有效地加快不确定性属性约简的运算速度。提出了可变截集变精度模糊粗糙集模型:综合地处理不确定性属性约简中的不确定性问题,可以通过改变等价类的截集来去除数据集噪声引起的不确定性;根据需要设定近似概念的精度,从而减少决策概念中含有的不确定性,使得到的决策结果更加可信。给出了一种伪核集粒子群搜索算法:可以减少大规模的数据集处理中花费的时间,满足实时性要求,该搜索算法将粒子的组成与约简的伪核集相联系,从而使粒子朝着更优的方向进化,能加速不确定性属性约简算法的收敛过程。研究了一种考虑不确定性的服装搭配算法:首次给出了基于不确定性属性约简和规则推理的服装搭配系统,能自动地找出服装搭配组合,克服了服装搭配与人的主观性因素造成的不确定性,并且减少了最终搭配方案的误分率。更多还原

【Abstract】 There lies some indeterminacy in the fields such as artificial intelligence, machine learning, pattern recognition and data mining, i.e randomicity, fuzzyness, and also imcompletion, inconsistency, time variability e.t.c. These indeterminacy must be cared on the data processing or analyzing, especially in the course of attribute reduction with the decision- making sort. So, we research a method serving the attribute reduction which is based on the fuzzy rough set and equavelant cluster in the environment with uncertainty, including its application on the fashion’s match, otherwise we wouldn’t obtain satisfied reduction object. While our main study contents and creative points are shown as the following:Giving out a sort of equavelant cluster based on the fuzzy rough set(FRS); using fuzzy c means method to repartition the universe of FRS to get new equavelant classes; these classes having smaller average cardinal number, able to accelerate the speed of convergence in the course of the reduction mentioned above.Putting forward a model of variable level and precision fuzzy rough set: processing uncertainty questions in the course of attribute reduction with indetermincy; removing those uncertainty aroused by noise through adapting the level of equavelant classes; reducing the uncertainty involved in the decision concepts after having set the precision of approximative notion according to the requisition in order to make decision result more reliable.Creating an algorithm of pudo-core-set particle optimisition: decreasing the time a lot spent in the processing of large-scale dataset, agreeing the demand of real time, the algorithm also relating the compisition of particles to pudo core set of reduction, thereby making particles evolving to an optimal direction, speeding up the convergence course of the algorithm.Studying an algorithm of fashion match, which considers uncertainty: primarily giving out a system of fashion match based on the attribute reduction with indeterminacy and rule reasioning, able to find out fashion match pattern automatically, to overcome the uncertainty resulted in the fashion match and human subjectivity factor, to eliminate the error rate of final match pattern.更多还原