Adaptive SVD de-noising method based on information entropy and application

Abstract

Wind turbine vibration signal is disturbed by noise, which seriously affects the subsequent fault diagnosis of the turbine. The existing de-noising methods have problems like de-noising principle not clear, operation not easy to carry out, and the de-noising effect not obvious. An adaptive SVD de-noising method based on entropy is put forward. Principle of adaptive SVD de-noising is analyzed, and two main methods of high dimension reconstruction are compared; High dimensional Hankel matrix is reconstructed according to the parity of length time series and the inequality principle and second order difference spectrum of information entropy is proposed to determine de-noising order. The proposed SVD de-noising method is applied in noise analysis of wind turbine vibration signals with comparison to three other SVD de-noising methods. Results show that the second order difference spectrum of information entropy SVD de-noising method has better smoothness, and it is capable to highlight the useful part of the signal, and achieves de-noising purpose effectively. The proposed de-noising method can provide reliable data basis for further fault diagnosis of turbines.