An Adaptive CEEMDAN Thresholding Denoising Method Optimized by Nonlocal Means Algorithm

A complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) thresholding denoising method optimized by nonlocal means (NLM) algorithm is proposed in this article. First, in order to enhance the adaptability and the accuracy of the algorithm, a composite screening method based on sample entropy–probability density–Mahalanobis distance for intrinsic mode functions (IMFs) is proposed. According to the proposed screening method, the IMFs are divided into three levels. Second, in order to obtain a threshold which can be adaptively changed, a threshold evaluation criterion is proposed to assist in selecting a suitable threshold. Then, the optimized thresholding denoising algorithm by the NLM is introduced to denoise the IMFs of different levels, in which the NLM algorithm with different parameters is used to smooth the different IMFs. Finally, all IMFs are reconstructed to obtain the denoised signal. The results of numerical simulation and experimental analysis to Doppler, Bumps, Signal3 (randomly generated nonstandard test signal) signals, partial discharge (PD) signals, and real signals show that the method of this article improves shortcomings of the traditional thresholding denoising method, such as inaccurate threshold selection, discontinuity of the data points of the denoised signals, and that the structure of the denoised signal is easily destroyed and the useful small-amplitude part of the denoised signal is easily discarded. The algorithm has better adaptability.