基于Gini-PSO-DPC算法的物联网数据异常检测  

Anomaly Detection of IoT Data Based on Gini-PSO-DPC Algorithm

作　　者：罗碧 沈艳 LUO Bi;SHEN Yan(School of Computer Science,Chengdu University of Information Technology,Chengdu 610000,China)

机构地区：[1]成都信息工程大学计算机学院,四川成都610000

出　　处：《软件导刊》2025年第2期98-106,共9页Software Guide

基　　金：国家自然科学基金项目(6217206);四川省科技计划重点研发项目(2023YFG0116)。

摘　　要：随着物联网设备的海量增长,其产生的数据也呈指数级增长。只有当数据具有可接受的质量时,其才具有价值,而海量数据中难免存在噪声数据。针对该问题,提出一种基于基尼系数与粒子群优化算法的密度峰值聚类算法Gini-PSO-DPC。首先,采用基于所有数据点的基尼系数计算最优截止距离dc;其次,通过粒子群优化算法寻找K个近似最优初始聚类中心,生成K个初始类别集群;最后,通过基于密度的最近数据点所属类别将样本数据点分配到对应类别集群。仿真实验结果表明,Gini-PSO-DPC算法的平均准确率达到96.81%,分别相较改进K-means、DMGA-FCM、DPC 3种算法提高了2.44%、0.89%、0.9%;平均精确率达到94.3%,分别相较改进K-means、DMGA-FCM、DPC 3种算法提高了1.22%、2.02%、1.33%。在消融实验中,Gini-PSO-DPC算法截止距离dc参数设置更加稳定合理,聚类时间更短,表明该算法具有更强的全局搜索能力、更高的自适应性和更好的聚类效果。With the massive growth of IoT devices,the data they generate is also growing exponentially.Data only has value when it has acceptable quality,and noisy data is inevitable in massive amounts of data.A density peak clustering algorithm called Gini PSO-DPC based on Gini coefficient and particle swarm optimization algorithm is proposed to address this issue.Firstly,the optimal cutoff distance is calculated using the Gini coefficient based on all data points;Secondly,the particle swarm optimization algorithm is used to find K approximately optimal initial cluster centers and generate K initial category clusters;Finally,the sample data points are assigned to the corresponding category cluster based on the density of the nearest data point's category.The simulation experiment results show that the average accuracy of the Gini PSODPC algorithm reaches 96.81%,which is 2.44%,0.89%,and 0.9% higher than the improved K-means,DMGA-FCM,and DPC algorithms,respectively;The average accuracy reached 94.3%,which was 1.22%,2.02%,and 1.33% higher than the improved K-means,DMGAFCM,and DPC algorithms,respectively.In the ablation experiment,the Gini PSO-DPC algorithm showed a more stable and reasonable cutoff distance parameter setting,shorter clustering time,indicating that the algorithm has stronger global search ability,higher adaptability,and better clustering effect.

关 键 词：物联网 聚类算法 DPC Gini-PSO-DPC 异常检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]