改进K-means结合深度学习的不完备信息选取  

Improved K-Means Combined with Incomplete Information Selection of Deep Learning

作　　者：宋新鹏[1] 张彦波[2] SONG Xin-peng;ZHANG Yan-bo(Office of Informatization Management,Henan University,Kaifeng Henan 475004,China;Shool of Physics and Electronics,Henan University,Kaifeng Henan 475004,China)

机构地区：[1]河南大学信息化管理办公室,河南开封475004 [2]河南大学物理与电子学院,河南开封475004

出　　处：《计算机仿真》2021年第9期433-437,共5页Computer Simulation

摘　　要：由于信息受到随机性和离群问题的干扰,导致信息聚类效果不好,从大量不完备数据中划分出有效信息成为各行各业所面临的重要问题。为此,提出改进K-means算法结合深度学习对不完备信息进行选取。首先,建立不完备信息系统粗糙集的表达形式,通过状态集、动作集对信息样本数据的三种行为进行决策,求得不完备信息系统粗糙集的上近似值和下近似值。其次,通过正同域、负反域和边界域来表示不完备信息的聚类结果,采用改进的K-means聚类算法,对缺失属性值进行集对分析。最后,通过深度学习对样本信息进行不断训练,求得这一过程的最优策略。基于MATLAB仿真实验,验证了在不完备信息选取过程中,所提算法能够获得正同域聚类结果以及正同域和边界域聚类结果的最优值,即使在一定的缺失率情况下,仍能保持较高的聚类效果,能够有效应用于不完备信息的选取。Due to the interference of randomness and outliers, the clustering effect of information is not good.It is an important problem for all walks of life to separate effective information from a large number of incomplete data.Therefore, an improved k-means algorithm combined with deep learning is proposed to select incomplete information.Firstly, the expression of rough sets of the incomplete information system was established.Through the state set and action set, the three behaviors of information sample data were determined.The upper and lower approximations of rough sets in incomplete information systems were obtained.Secondly, the clustering results of incomplete information were represented by positive region, negative inverse region and boundary region.The improved k-means clustering algorithm was used to analyze the missing attribute values.Finally, the optimal strategy of this process was obtained by continuous training of sample information through deep learning.Based on the MATLAB simulation experiment, it is verified that in the process of incomplete information selection, the proposed algorithm can obtain positive region clustering results the optimal values of the clustering results in the positive region and the boundary region.Even in the case of a certain missing rate, it can still maintain a high clustering effect, and it can be effectively applied to the selection of incomplete information.

关 键 词：不完备信息 粗糙集 深度学习 最优策略 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]