基于粒度理论的高维数据流并行计算方法  

Parallel Computing Method for High Dimensional Data Flow Based on Granularity Theory

作　　者：路晶[1] 胡顺仿 LU Jing;HU Shun-fang(China Civil Aviation Flight Academy,Guanghan Sichuan 618307,China;Yunnan National University,Kunming Yunnan 650000,China)

机构地区：[1]中国民用航空飞行学院,四川广汉618307 [2]云南民族大学,云南昆明650000

出　　处：《计算机仿真》2021年第5期246-249,422,共5页Computer Simulation

基　　金：民航安全能力基金项目(省部级)(ASSA2019/20);民航飞行技术与飞行安全重点实验室自主研究项目(校级)(FZ2020ZZ02)。

摘　　要：以实现多种形态高维数据流的高效、精确并行计算为出发点,提出基于粒度理论的高维数据流并行计算方法。使用基于动态粒度的数据流挖掘模型,高效挖掘高维数据流;利用基于局部保持投影原理和主成分分析原理压制高维数据流噪声,减少高维数据流噪声隐患;依据降噪后不同高维数据流特点,采用高维数据流相关性分析并行计算方法,得到高维数据的皮尔逊积差相关系数,实现数据流关联,并基于数据流十字转门模型,定义适合高维数据流分析的滑动数据流窗口模式,实现高维数据流的并行计算。实验结果验证,上述方法挖掘高维数据流的内存消耗低,高维数据流数据去噪能力强,具备较高的高维数据流并行计算精度,且并行计算效率高。This paper proposes a parallel computing method of high-dimensional data streams based on granularity theory for achieving efficient and accurate parallel computing of multi-dimensional data streams. The data stream mining model based on dynamic granularity was adopted to mine high-dimensional data stream efficiently. The noise of high-dimensional data stream was compressed based on the principles of locality preserving projection and major constituent analysis, lowing the hidden trouble of high-dimensional data stream noise. Based on the characteristics of different high-dimensional data streams after noise reduction, the parallel computing method was analyzed in detail to get the Pearson product difference correlation coefficient of high-dimensional data via the correlation of high-dimensional data streams, realizing the data stream correlation. According to the data flow turnstile model, the sliding data flow window mode suitable for high-dimensional data flow analysis was defined, and the parallel computing of high-dimensional data flow was finally achieved. The results show that the proposed method has low memory consumption, excellent denoising ability, high parallel computing accuracy and efficiency.

关 键 词：粒度理论 动态粒度 高维数据流 皮尔逊积差 并行计算 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]