The EMD Processing And BP Predicting Of Sensor Signal

The Empirical Mode Decomposition (EMD) method proposed by doctor E.Huang, whichis a new time-frequency analysis method is very suitable for nonlinear and non-stationarysignal processing.EMD algorithm is a new time-frequency analysis method, and it is very suitable fornonlinear, non-stationary signal, which is introduced into the prediction of the sensor signal,the EMD processing, nonlinear, non-stationary sensor signal is decomposed into relativelystable IMF component in this dissertation. Then the neural network to predict the IMFcomponent, the predicted results together to obtain the prediction results of the sensor signal.Firstly we introduce the theory of the time-frequency analysis of traditional fields, andthen explains the limitations of traditional methods, thus the introduction of Hilbert Huangtransform (Hilbert-Huang Transform) and its core part of empirical mode decomposition areachieved.In introducing the EMD process, we mainly focus on these parts: sampling rate, endeffect, interpolation, mode mixing and the stop criteria of the algorithm. By comparing themirror method, we introduce the improved slope method to limit the end effect. We useFrench scholar G. Rilling’s threshold method as the sifting conditions, then we decompose theoriginal signal into some intrinsic mode functions (IMF).Then we discuss the performance of some neural network, and compare the advantagesand disadvantages of N-1and N-M method in prediction. In the simulation experiment, weapply back-propagation (BP) neural network and the N-1method into predicting. Thenpredictive value are added as the forecasting value of the original signal. At the same time weuse BP network to directly predict the signal without decomposition.Finally, the comparison of the two predicting results is achieved。Then we can concludethat for first non-stationary signals, the result of prediction of signal with empirical modedecomposition is better that without empirical mode decomposition by directly applying BPprediction to the signal.