Fast Least Squares Multivariate Empirical Mode Decomposition for Plant-Wide Oscillation Extraction

The plant-wide oscillation in process industry can lead to problems such as high scrap rate, high energy consumption, and reduced stability of machine operation. Aiming at the problem that existing decomposition methods are difficult to extract the oscillatory components efficiently and completely, this work proposes a fast least squares multivariate empirical mode decomposition (FLSMEMD). The method performs quantitative screening of the projected sequences according to the number of extreme points and local fluctuation characteristics, and then extracts the intrinsic mode functions from the resulting sequences. Finally, the output of FLSMEMD is derived by solving an overdetermined linear equation system. Experimental results on simulated signals and real industrial cases demonstrate that FLSMEMD is capable of effectively suppressing mode mixing and distortion. In addition, it overcomes the computational inefficiency of the decomposition process due to the redundancy of projection directions.