基于云平台下的并行VMD算法

Parallel Variational Mode Decomposition Based on Cloud Platform

电力设备监测数据中的时序波形信号对评估设备运行状态具有重要作用,当数据量很大时,采用传统方法处理时序波形信号往往效率低下。VMD算法是一种完全非递归的变分模态分解方法,适用于对非线性、非平稳信号的分析,但其复杂性和大量的计算限制了其应用范围。提出了一种基于Storm/Spark平台的并行VMD算法。为确保子段数据模态分量在窗口截断处连续,提出了一种基于矩形窗分段的变分模态分解(Variational Mode Decomposition,VMD)VMD-RWS和基于自适应分段和外推延拓(the adaptive subsection based on local flatness and extrapolation,ASLF-E)ASLF-E的信号处理方法进行信号分段以及子段数据处理。通过仿真实验对提出的VMD-RWS和ASLF-E方法进行验证,实验结果表明该方法可以确保各子段数据模态分量在窗口截断处的连续性,保持了VMD算法原有的性能,为云平台下局部放电信号应用VMD算法进行分析提供了一种切实可行的方案。此外,设计了基于Storm和Spark平台的并行VMD算法处理流程及架构,为基于云平台的并行VMD算法的实现提供了依据。

The waveform signals in monitoring data of power equipment play an important role to evaluate the state of power equipments.When data volume is large,traditional methods are often inefficient in processing the waveform signals.As a completely non-recursive variational mode decomposition method,VMD algorithm is applicable to the analysis of non-linear and non-stationary signals.However,its application is limited due to complexity and large amount of calculation.To ensure the continuity of sub modal data components at the window boundary,this paper proposed a parallel VMD algorithm based on rectangular window segmentation(VMD-RWS)and an adaptive subsection based on local flatness and extrapolation(ASLF-E)to process signal segmentation and sub segment data.The VMD-RWS analysis method based on ASLF-E was validated by simulation experiment.The experiment results demonstrate that the proposed method ensures the continuity of each sub segment data modal components at window boundary.It maintains the original performance of the VMD algorithm.It provides a feasible scheme for analyzing partial discharge signals on cloud platform by using VMD algorithm.In addition,this paper designs a parallel VMD algorithm to process flow and architecture based on Storm and Spark platform,and provides basis for the implementation of parallel VMD algorithm on cloud platform.