间歇性时间序列数据多目标决策挖掘算法设计  

Design of Multi-objective Decision Mining Algorithm for Intermittent Time Series Data

作　　者：张伟[1] 刘新[2] ZhANG Wei;LIU Xin(Huaiyin Institute of Technology Huai'an,Jiangsu 223003,China;School of Business,Central South University of China,Changsha Hunan 410083,China)

机构地区：[1]淮阴工学院,江苏淮安223003 [2]中南大学商学院,湖南长沙410083

出　　处：《计算机仿真》2024年第10期291-295,300,共6页Computer Simulation

基　　金：江苏省现代教育技术研究2022年智慧校园专项立项课题(2022-R-107214)。

摘　　要：时序数据规模化存储能大幅度提升经济效益,但传统挖掘算法无法从海量数据中提取有效信息。为解决上述难题,通过对数据进行优化处理,提出一种MAD-SVR时序数据回归预测算法。算法首先对大数据进行标准化处理,并通过极值点分析与剔除,提升数据的有效性;然后采用多目标MIC相关性分析方法,提高对标准时序数据的间歇性特征提取能力;接着利用AHP层次分析量化指标,获取最优簇N,并基于DIANA算法完成时序数据聚类优化过程;最后通过十折交叉验证的方式,构建SVR时序数据回归预测模型,完成预测结果输出。不同叠加模型的仿真对比结果表明,较其他模型相比,MADSVR模型的MAPE参数整体减少了53.89%,R^(2)参数增加了5.99%,且RMSE参数至少下降了12.31%,即其该模型的拟合度最高,预测能力最优,且预测真实占比有较大提高,但预测真实占比误差偏离度尚有优化空间。综上,MAD-SVR算法在海量时序数据挖掘中具有重要的仿真研究价值。Large-scale storage of time series data can greatly improve economic benefits,but traditional mining algorithms cannot extract effective information from massive data.In order to solve the above problems,this paper proposes a MAD-SVR regression prediction algorithm for time series data by optimizing the data.Firstly,the algorithm standardized the large data,and improves the validity of the data through the analysis and elimination of extreme points,and then used the multi-objective MIC correlation analysis method to improve the intermittent feature extraction ability of the standard time series data;Then,the optimal cluster number N was obtained by using AHP to analyze the quantitative indicators,and the clustering optimization process of time series data was completed based on DIANA algorithm.Finally,the SVR regression prediction model of time series data was constructed by ten-fold cross validation,and the prediction results were output.The simulation results of different superposition models show that,compared with other models,the MAPE parameters of MAD-SVR model are reduced by 53.89%,the R^(2) parameters are increased by 5.99%,and the RMSE parameters are reduced by at least 12.31%,that is,the model has the highest fitting degree,the most precise prediction ability,and the proportion of real prediction is greatly improved.However there is still room for optimization in predicting the deviation of the real proportion error.To sum up,MAD-SVR algorithm has important simulation research value in massive time series data mining.

关 键 词：时间序列数据 多目标分析 回归模型 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]