基于二进制区分矩阵的增量式知识约简算法研究

发布时间：2018-04-10 21:25

本文选题：粗糙集 + 二进制区分矩阵　；参考：《南京邮电大学》2017年硕士论文

【摘要】：知识约简是粗糙集理论的核心内容之一。通过知识约简可以在保证信息系统决策和分类能力不变的前提下剔除数据集中冗余信息。现实生活中,数据以不可预期的速度在增加。每获得一个新对象数据,在冗余信息剔除计算中都对整个数据集重新进行知识约简计算,必然是浪费时间和低效的。因此,对于以原有决策表知识约简计算结果为基础,计算新增加部分从而获得新决策表知识约简的增量式知识约简算法具有重要的理论和现实意义。本文针对传统二进制区分矩阵存储空间大以及如何有效地将二进制矩阵在完备和不完备信息系统中用于增量式知识约简的问题,研究了基于二进制区分矩阵的增量式知识约简算法,并将约简算法用于光伏发电功率预测系统的数据预处理,主要研究内容包括:(1)探索了在完备信息系统下基于二进制区分矩阵的增量式属性约简算法。为了解决二进制区分矩阵存储空间大的问题,提出了一种压缩二进制区分矩阵的方法,将二进制区分矩阵的存储空间从|C|+1列简化成3列。当增加单个新实例时,根据建立的压缩二进制区分矩阵,通过动态更新二进制区分矩阵的方法实现增量式属性求核,并以属性核为出发点,提出了在增加单个实例时基于二进制区分矩阵的增量式属性约简算法。(2)探索了在完备信息系统下增加成组数据时基于二进制区分矩阵的增量式属性约简算法。根据新增数据是单一实例还是成组实例对象,选择不同的分支程序来更新二进制区分矩阵。根据更新后的二进制区分矩阵快速求核,并以属性核为出发点,提出了适用于成组对象增加的基于二进制区分矩阵的增量式属性约简算法。(3)探索了基于二进制区分矩阵的不完备信息系统增量式属性约简算法。不完备信息系统下的增量式属性约简求解首先需要求解容差类。当在已有系统中新增实例时,为了快速求解新的容差类,首先提出了一种快速、稳定性较好的容差类静态求解方法,然后在此基础上提出了容差类的增量式求解方法。根据增量式求得的新容差类,通过动态更新二进制区分矩阵,提出了不完备信息系统下基于二进制区分矩阵的增量式属性约简算法。(4)探索了增量式属性约简算法用于光伏发电功率预测数据的特征提取。对采集的光伏数据建立光伏发电功率预测数据模型决策表,并对采集到的光伏数据进行相应的离散化处理。当新增数据时采用增量式属性约简算法进行特征提取,并对提取特征数据采用神经网络算法进行训练和预测。
[Abstract]:Knowledge reduction is one of the core contents of rough set theory.The redundant information in the data set can be eliminated by knowledge reduction on the premise that the decision and classification ability of the information system is invariable.In real life, data is increasing at an unexpected rate.It is a waste of time and inefficiency to recompute the knowledge reduction of the whole data set in the computation of redundant information elimination for each new object data.Therefore, it is of great theoretical and practical significance for the incremental knowledge reduction algorithm to obtain the new decision table knowledge reduction based on the results of the original decision table knowledge reduction.This paper aims at the problem of large storage space of traditional binary discernibility matrix and how to effectively apply binary matrix to incremental knowledge reduction in complete and incomplete information systems.The incremental knowledge reduction algorithm based on binary discernibility matrix is studied, and the reduction algorithm is applied to the data preprocessing of photovoltaic power prediction system.The main research contents include: 1) the incremental attribute reduction algorithm based on binary discernibility matrix in complete information system is explored.In order to solve the problem of large storage space of binary discernibility matrix, a method of compressing binary discernibility matrix is proposed. The storage space of binary discernibility matrix is simplified from C1 column to 3 column.When a single new instance is added, according to the compressed binary discriminant matrix, the incremental attribute kernel is realized by dynamically updating the binary discernibility matrix, and the starting point is attribute kernel.An incremental attribute reduction algorithm based on binary discernibility matrix is proposed when adding a single instance. The incremental attribute reduction algorithm based on binary discernibility matrix is explored when adding group data in a complete information system.According to whether the new data is a single instance or a group instance object, different branch programs are selected to update the binary discriminant matrix.Based on the updated binary discernibility matrix, the kernel is quickly obtained and the attribute kernel is used as the starting point.An incremental attribute reduction algorithm based on binary discernibility matrix is proposed for adding groups of objects. The incremental attribute reduction algorithm for incomplete information systems based on binary discernibility matrix is explored.In order to solve incremental attribute reduction in incomplete information systems, tolerance classes should be solved first.In order to solve the new tolerance class quickly, a fast and stable static solution method of tolerance class is proposed, and then an incremental solution method of tolerance class is proposed.According to the new tolerance class obtained by the increment formula, the binary discriminant matrix is dynamically updated.An incremental attribute reduction algorithm based on binary discernibility matrix in incomplete information systems is proposed. The incremental attribute reduction algorithm is explored for feature extraction of photovoltaic power prediction data.The model decision table of photovoltaic power prediction data is established for the collected photovoltaic data, and the corresponding discrete processing of the collected photovoltaic data is carried out.When new data is added, incremental attribute reduction algorithm is used for feature extraction, and neural network algorithm is used to train and predict feature data.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】