基于HMM校正与神经网络延拓的EMD端点效应抑制方法

针对神经网络延拓方法在抑制经验模态分解的端点效应时存在的延拓数据与真实数据往往存在误差的问题，提出了一种基于HMM校正的方法来减小预测延拓数据误差。首先利用径向基函数(RBF)神经网络预测估计方法对部分原始数据进行估计，同时对端点外数据进行预测。然后计算该方法估计的数据与真实数据的误差序列,再用HMM方法建立估计误差序列模型，用以预测延拓后数据的误差。最后用RBF神经网络延拓数据减去HMM预测的误差数据得到新的校正后延拓数据。仿真与实验证明了将HMM预测方法与RBF神经网络数据延拓结合应用到解决端点效应的过程中所得到的延拓数据更接近真实数据，能够更好地解决端点效应问题，提高了经验模态分解精度。

Restraining Method of End Effect for EDM Based on Error Calibration by HMM and Neural Network

End effects reduced the precision of EMD greatly, and neural network extension was used to restrain the end effects. However, there were forecast errors in the data forecasted by neural network. Here, a new method, error calibration by HMM was proposed to solve the problem. The radial basis function(RBF) neural network was firstly used to forecast outboard of both ends and estimation part of original signals. Then HMM was used to analyze the forecasting errors and to find the regularity of forecast errors of neural network. According to the analysis, the forecasting errors in next step were analyzed and forecasted to adjust neural network's forecasting results. Simulation and experiments prove that the method can solve the end effects effectively.