A Variational Mode Decomposition Algorithm Based on Orthogonal Criterion for Robust Feature Extraction

Variational mode decomposition (VMD) is one of the most powerful adaptive mode decomposition (AMD) methods. However, its performance highly depends on the initialization of several parameters. In this paper, we propose an extension of VMD, namely the empirical spectral trend guided adaptive orthogonality-constrained VMD (EST-AOCVMD). Our method determines the number of modes, initial center frequencies and bandwidth based on the proposed EST method, which incorporates the noise background normalization to improve its accuracy in detecting spectral fluctuation affected by noise. The mode decomposition is accomplished by introducing an orthogonal criterion into an optimization problem that addresses the mode mixing issue. The bandwidth of each mode is also adaptively updated based on the orthogonality between modes. The performance analysis suggests that the EST is capable of dealing with noise interference and provides a robust guide for the decomposition of multicomponent signals without the need for prior knowledge, which in turn contributes to computational efficiency. Moreover, the simulation results and experimental data validations demonstrate that the proposed EST-AOCVMD exhibits clear advantages over some classic AMD methods and is promising for a broad set of applications concerning nonstationary signal processing and feature extraction in noisy environments.