摘要
多元信息流涵盖多种类型、不同维度的数据,存在异构性和不确定性,以大规模和高速度连续产生,难以从中提取异常数据分布的特征,使得算法对异常数据的检测陷入局部最优解,出现局部收敛和早熟现象,进而影响异常数据修正。对此,引入混合蛙跳算法,对多元信息流异常数据展开修正。标准化处理多元信息流数据,建立预选特征子集,采用混合蛙跳算法-人工鱼群算法和混合蛙跳算法-模糊C-均值聚类算法,在搜索过程中利用了两种算法的优势,在多个搜索空间中找到最优数据特征,更准确地划分聚类簇,获取最优的数据特征并实施聚类处理,得到异常数据集合。基于单层前馈神经网络,构建异常数据修正模型,通过更新参数,由输出层输出异常数据的修正结果。仿真测试结果显示:混合蛙跳算法能够加强融合对象的优势,检测异常数据集占比高达99.72%,精准完成异常数据检测任务;修正误差最大仅为1.119,可以满足精准性需求。
The multi-dimensional information flow covers various types of data with different dimensions,which is heterogeneous and uncertain.It is difficult to extract the characteristics of abnormal data distribution from it,which makes the algorithm fall into local optimal solution,local convergence and premature phenomenon,and then affects the correction of abnormal data.In this paper,the mixed Shufled Frog Leading Algorithm is introduced to modify the abnormal data of multi-information flow.Standardize the multi-information flow data,establish a pre-selected feature subset,and adopt the hybrid Shufled Frog Leading Algorithm-Artificial Fish Swarm Algorithm and the hybrid Shuffled Frog Leading Algorithm-Fuzzy C-means clustering algorithm.In the search process,the advantages of the two algorithms are used to find the optimal data features in multiple search spaces,divide the clustering clusters more accurately,obtain the optimal data features and implement clustering processing to get the abnormal data set.Based on single-layer feedforward neural network,the abnormal data correction model is constructed,and the correction results of abnormal data are output by the output layer by updating parameters.The simulation test results show that the hybrid Shuffled Frog Leading Algorithm can strengthen the advantages of the fusion object,and the proportion of abnormal data sets detected is as high as 99.72%,thus completing the task of abnormal data detection accurately.The maximum correction error is only 1.119,which can meet the requirement of accuracy.
作者
颜清
李金讯
陈诗
YAN Qing;LI Jin-xun;CHEN Shi(Hainan Power Grid Co.,Ltd.Information and Communication Branch,Haikou Hainan 570000,China)
出处
《计算机仿真》
2025年第1期258-262,共5页
Computer Simulation
基金
海南电网有限责任公司项目(202000123202)。
关键词
多元信息流
混合蛙跳算法
人工鱼群算法
模糊均值聚类算法
单层前馈神经网络
异常数据修正
Multiple information flow
Mixed Shuffled Frog Leading Algorithm
Artificial fish swarm algorithm
Fuzzy mean clustering algorithm
Single-layer feedforward neural network
Abnormal data correction
