Abstract: In order to solve the problem of extracting harmonic parameters by noise interference Prony

algorithm，a harmonic analysis method based on ensemble local mean decomposition（ELMD），singular value

decomposition（SVD）and Prony was proposed. Firstly， the noisy signal was decomposed by ELMD，and the

demarcation point between the noisy component and the effective component was determined by Kullback-Leibler

divergence for a series of product functions（PF）. The noise component was removed and the effective component

was retained. The Hankel matrix was reconstructed through the phase space for the effective component，and SVD

was used for secondary noise reduction and reconstruction. Finally，the reconstructed signal was superimposed with

the ELMD remainder to get the de-noised harmonic signal，which was combined with the Prony algorithm to detect

the frequency，amplitude and phase of the harmonic. It shows that this method can effectively reduce noise and

extract harmonic characteristic parameters through the simulation experiment results.

      Key words: ensemble local mean decomposition（ELMD）；singular value decomposition（SVD）；Prony

algorithm；Kullback-Leibler divergence；denoising；harmonic detection