Abstract: Given the fact that hydroelectric generating units are often used for peak and frequency modulation and spinning reserve, noise reduction of their vibration signals is dramatically significant to promoting the incipient fault identification and safe operation of power systems. This paper develops a novel EMD-based denoising algorithm using the similarity measure between consecutive geometric distributions. The signals was reconstructed by using different intrinsic mode functions generated from EMD sifting, and fitted the probability density functions of these reconstructed signals by the nonparametric kernel density estimation theory. Then, a Hausdorff distance was adopted to calculate the indexes for evaluating the similarity measure between the consecutive geometric distributions of probability density functions, and an optimal separation between characteristic IMF components and noisy IMF components is carried out through variation trend analysis of the similarity measure indexes. This method is validated using model simulations and engineering application, and the results demonstrate it achieves a remarkable effect on noise reduction of hydroelectric generating unit signals.

Key words: hydroelectric generating units, signal noise reduction, empirical mode decomposition, geometric distribution similarity, Hausdorff distance