【作者】 杨洋；

【导师】 武妍；

【作者基本信息】 同济大学 ， 模式识别与智能系统， 2006， 硕士

【摘要】 粗糙集理论是一种新颖、有效的软科学计算方法,能够分析和处理不完备信息。本文着重对粗糙集理论的基本问题——连续属性离散化、决策表属性约简,进行了深入的研究。分析了遗传算法和粗糙集理论的特点以及二者结合的优势,进而研究了神经网络模型参数的确定方法,提出了基于粗糙集的神经网络模型的构建方法。 本文的研究工作包括有以下几个方面: 第一,对连续属性离散化方法进行了研究,提出了基于遗传算法和粗糙集理论的连续属性离散化算法。该算法在保持原始属性分类能力不变的情况下,对原始所有断点进行全局快速寻优,获得属性的最小离散集。通过实验证明了算法在连续属性离散化方面的有效性。 第二,对粗糙集属性约简方法进行了研究,提出了基于遗传算法和粗糙集理论的属性约简方法,即应用遗传算法对粗糙集简化形成的知识表进行全局寻优,获得最小属性集。该算法既具有粗糙集处理不一致训练数据的能力,又具有遗传算法有效搜索决策属性的能力。实验表明,加入遗传搜索器的粗糙集产生的属性集具有简化、实用的特点,缩短了计算时间,因此,将两种算法结合具有可行性。通过实验证明了粗糙集理论和遗传算法结合在属性约简方面的有效性。 第三,研究了BP神经网络模型的构成及特点,针对神经网络模型的不足,论述了粗糙集与神经网络结合的可能性。利用粗糙集在处理不确定性和不完全性问题方面的优点,构成粗糙集神经网络模型,即根据基于粗糙集和遗传算法获取的最优数据集,提取最简规则构建新的神经网络模型。仿真实验研究表明,采用基于粗糙集和遗传算法的神经网络不仅保持了原有神经网络模型和粗糙集理论的优点,还在实时性方面获得了较大的提高。 最后,对全文所做的工作进行了总结,并对下一步研究工作进行了展望。更多还原

【Abstract】 Rough set theory has emerged as a major mathematical method to manage uncertainty which is from inexact, noisy and incomplete information. The thesis places emphasis on study of the basic problems about rough set theory—real-value attributes quantization and attributes reduction. The research about integrating genetic algorithm with rough set theory is deep and effective. Using the results concluded from the combined algorithms determining neural networks parameters can make up the shortages of neural networks. The constructed method of neural networks based on genetic algorithm and rough set method is proposed.The main contents are shown as follows.Firstly, the method of real-value attributes quantization is studied and another method based rough set theory and genetic algorithm is brought forward. The new method is the hybrid of these two intelligent algorithms. The experiment result has successfully shown that it is possible to integrate rough set theory with a GA-based search algorithm to perform quantization group from examples with continuous data. The quantization group of attributes produced is simple and concise, and the training time decreases greatly.Secondly, the methods of attributes reduction using rough set theory are introduced and another method based rough set theory and genetic algorithm is brought forward. Rough set makes the data into an information system, and then genetic algorithm begins to look for the reduction attributes group. Rough set theory provides a novel way of dealing with vagueness and uncertainty. When coupled with genetic algorithm, the reduction attributes group is able to be obtained from inconsistent information. The new method has these merits. The experiment results have successfully shown that it is possible to integrate rough set theory with a GA-based search algorithm to perform attributes reduction efficiently.Thirdly, the structure and the traits of BP neural network are researched. The possibility of the integration of rough set method and neural network based theshortage of BP network is discussed. The advantages of rough set theory, in dealing with the problems of uncertainties and incompleteness, are been deep into study. Another new method based on the rough set theory, genetic algorithm and neural network is worked out, and the architecture, the algorithm, and the performances of this method are researched and analyzed in detailed, a large number of simulative experiments are carried on. And as the results show, some satisfied performances are given by the method of neural network based on rough sets, such as the performance in dealing with the problem of real-time. Some key difficulties of theory and technique of the method, which mentioned in the thesis, are solved well. In fact, the hybrid of three intelligent algorithms is great significance.Finally, the summary of the whole work is given, while some of unsolved problems in the thesis and the prospect of the further study are indicated.更多还原