基于K-近邻与FOA改进聚类的数据异常分析模型及用电行为分析  

Data Anomaly Analysis Model and Electricity Consumption Behavior Analysis based on K-nearest Neighbor and FOA Improved Clustering

作　　者：周伟[1] 牛誉蓉 ZHOU Wei;NIU Yurong(Academic Affairs Office,Anhui Communications Vocational and Technical College,Hefei 230051,China;College of Computer Engineering,Anhui Wenda University of Information Engineering,Hefei 231201,China)

机构地区：[1]安徽交通职业技术学院教务处,合肥230051 [2]安徽文达信息工程学院计算机工程学院,合肥231201

出　　处：《成都工业学院学报》2024年第5期11-16,共6页Journal of Chengdu Technological University

基　　金：安徽省高等学校自然科学研究项目(2023AH052967)。

摘　　要：对隐藏在大数据中的信息进行深层挖掘时,由于存在数据来源、统计口径、人员输入、行为异常等方面的问题,可能出现异常数据。针对此类问题,首先利用离散小波变换进行多尺度分解,然后采用K-近邻思想对局部区域的密度、距离重新定义,来提高对异常值的识别精度;最后结合改进的果蝇优化算法,对密度峰值聚类算法中的截断距离进行优化,提出基于K-近邻与改进果蝇优化的密度峰值聚类异常分析模型。从异常值检测角度进行仿真实验分析,根据用户数据多时间尺度特征,对不同时间尺度的复合数据进行聚类,对用电行为进行分析;选择多种标准测试函数,对基于知识学习的改进果蝇优化算法性能进行对比研究。结果显示,基于K-近邻的算法能够将变压器中不同于正常运行模式的少数异常曲线及单个用户的异常用电模式检测出来,其有效性得到了验证。在基于知识学习的改进果蝇优化算法中,随着果蝇个体数量增加其寻优能力也得到提高。In the background of big data,when deep mining the information hidden in big data,there may be abnormal data due to the problems of data sources,statistical caliber,personnel input,behavioral anomalies and other aspects.To solve this problem,firstly,discrete wavelet transform was used for multi-scale decomposition,and then K-nearest neighbor idea was used to redefine the density and distance of local regions to improve the recognition accuracy of outliers.Finally,combined with the improved fruit fly optimization algorithm,the truncation distance of the density peak clustering algorithm was optimized,and the density peak clustering anomaly analysis model based on K-nearest neighbor and improved fruit fly optimization was proposed.From the perspective of outlier detection,the simulation experiment analysis was carried out.According to the multi-time scale characteristics of user data,the composite data of different time scales were clustered to analyze the electricity consumption behavior.The performance of improved fruit fly optimization algorithm based on knowledge learning was compared by selecting several standard test functions.The results show that the algorithm based on K-nearest neighbor can detect the few abnormal curves in the transformer that are different from the normal operation mode and the abnormal power consumption mode of a single user,and its effectiveness has been verified.In the improved fruit fly optimization algorithm based on knowledge learning,the optimization ability of fruit fly increases with the increase of individual number.

关 键 词：异常值检测 果蝇优化算法 K-近邻算法 峰值聚类算法 用电行为 

分 类 号：C931.1[经济管理—管理学]