用于厂级振荡提取的快速最小二乘多元经验模态分解方法

Afast least squares multivariate empirical modal decomposition method for plant-wide oscillation extraction

中文关键词  经验模态分解  波动特征  主成分分析  最小二乘

英文关键词  empirical mode decomposition  fluctuation characteristics  principal component analysis  least squares

中文摘要

      工业控制过程中潜在的厂级振荡行为会导致高废品率、高能耗、降低设备稳定性等问题.为此,本文提出 了一种快速最小二乘多元经验模态分解方法,用以高效提取过程输出数据中的厂级振荡分量.该方法首先按照极 值点数目和局部波动特征对投影序列作量化筛选;然后,使用主成分分析得到最能代表信号特征的投影序列;最后, 对得到的序列提取本征模态函数以构建超定线性方程组,并利用最小二乘法求解得到分解结果.仿真信号和实际工 业案例的实验结果表明,所提方法有效抑制了模态混叠、失真等问题,同时克服了分解过程中由于计算冗余投影方 向而造成的无效耗时问题.

英文摘要

      Plant-wide oscillation in process industry can lead to problems such as high scrap rate, high energy consump tion and reduced stability of machine operation. To facilitate more efficient and fast extraction of plant-wide oscillatory components from plant data, a fast least squares multivariate empirical modal decomposition (FLSMEMD) algorithm is proposed. The method first quantitatively filters the projection sequences based on the number of extremes and local fluc tuation characteristics. Then, principal component analysis is used to further separate the projection sequences that best represent the signal features. Finally, the intrinsic mode functions are extracted from the obtained sequences, and the output of FLSMEMD is derived by solving an overdetermined linear equation system through the use of least squares. Experi mental results on simulated signals and real industrial cases demonstrate that FLSMEMD is able to effectively suppress mode mixing and mode distortion. In addition, the proposed method overcomes the problem of inefficient computation due to redundant projection directions during the decomposition process.