Research On End Effect Of Decomposing Signal Into IMFs And Prediction Of Signal

In recent years, as a method of non-stationary signal processing in modern signalfield, HHT has been favored by more and more scholars. This method can analyzesignal in both time and frequency domain, so it can be well used to study non-linearand non-stationary signals, and it has been widely used in many fields such as ocean,vibration engineering, fault diagnosis, biomedical, financial, and so on, and hasachieved good results.EMD is a key part of HHT. Any complicated signal can be decomposed into afinite number of IMFs which are successively from high to low frequency. Theintroduction of IMF gives actual physical significance to the instantaneous frequency.EMD is a decomposition method based on time domain. The main work of this paperis based on EMD process.Firstly, end effect which has always been a problem of great concern in EMD isstudied. A method of boundary extension based on similar inner series of the signalwas proposed. In this method, the boundary of the signal has been extended morereasonably.Secondly, on the basis of FFDSI which is proposed by Zhang Lizhen, it improvesbetter the impact of end effect on EMD. In this method, the original signal wasdecomposed into IMFs by FFDSI at first. Then two appropriate sections interceptedfrom the interior were joined to the two sides of the original signal respectively. Bymeans of using FFDSI on the extended signal, decomposed results were obtained andthen the extensions of these IMFs were cut off in order to retain the decomposedresults have the same length as the original signal. By this method, the extendedsignal can well reflect the tendency of signal so it is more close to the real sequence. Besides, this method can get rid of the constraint from envelops which is a bigproblem in traditional extension. The experiment results show the method is effective.Thirdly, for the prediction of the signal, forecasting model based on the empiricalmode decomposition domain is proposed. By EMD, the signal can be decomposed asthe sum of several IMFs and a residual term, so the problem of the prediction for thesignal is transformed into the prediction of the IMFs and the residual term. Methodsbased on corrected trigonometric function and local quadratic function is used toforecast IMFs and residual term respectively. Then the prediction of the originalsignal is obtained. Since EMD can effectively retain the structure of signal, it is ofhigh accuracy. The experiment results show this method for prediction is effective.