基于HMM校正与神经网络延拓的EMD端点效应抑制方法  

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

作　　者：孟宗[1,2] 闫晓丽[1] 王赛[3] 

机构地区：[1]河北省测试计量技术及仪器重点实验室(燕山大学),秦皇岛066004 [2]国家冷轧板带装备及工艺工程技术研究中心,秦皇岛066004 [3]长城汽车股份有限公司技术中心(河北省汽车工程技术研究中心),保定071000

出　　处：《中国机械工程》2015年第14期1920-1925,共6页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(51105323);河北省自然科学基金资助项目(E2015203356);河北省高等学校科学研究计划资助重点项目(ZD2015049)

摘　　要：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。End effects reduced the precision of EMD greatly, and neural network extenslon 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 esti- mation 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.

关 键 词：隐马尔科夫模型 误差校正 神经网络 端点效应 经验模态分解 

分 类 号：TN911.7[电子电信—通信与信息系统] TH165[电子电信—信息与通信工程]