A quantified self-adaptive filtering method: effective IMFs selection based on CEEMD

Abstract

In the paper, a quantified self-adaptive signal-filtering method called effective intrinsic mode functions (IMFs) selection based on complementary ensemble empirical mode decomposition (CEEMD) is proposed. In the method, a combination of the self-adaptive separation of IMFs and the correlation analysis is presented. Generating IMFs by CEEMD are divided automatically into the noisy domain and the signal domain through the quantified correlation coefficient estimation. In order to choose effective IMFs, low-frequency priority and energy priority are used in the noisy domain and the signal domain respectively. A correlation threshold value is also quantitatively calculated by mutual information (MI) between adjacent IMFs. The threshold can filter out accurately IMFs that contain a lot of noises. The final reconstructed signal will be gotten via the partial reconstruction of effective IMFs selected in different domains. All parameters in the proposed filtering method are self-adaptively obtained without a priori knowledge. This method is a fully data-driven approach. Test results of simulation and real signals show the validity of the proposed method and demonstrate its superior performance compared with the other methods of references. In addition, the application of the proposed method and comparison experiment with CEEMDAN are also investigated. The study is limited to signals that were corrupted by additive white Gaussian noise.