基于自适应本征维数估计流形学习的相空间重构降噪方法

Phase space reconstruction method for vibration signal's noise reduction based on adaptive intrinsic dimension estimation manifold learning

针对实际工程领域振动信号噪声干扰大、具有强烈非线性等问题,提出了基于自适应本征维数估计流形学习的相空间重构降噪方法。利用相空间重构将一维含噪时间序列重构到高维相空间;基于极大似然估计法(maximum likelihood estimate,MLE)估计相空间中每个样本点的本征维数并使用自适应加权平均法计算全局本征维数;采用局部切空间排列(Local tangent space Alignment,LTSA)流形学习方法将含噪信号从高维相空间投影到有用信号的本征维空间中,剔除分布在高维空间中的噪声后,重构回一维时间序列。通过Lorenz仿真实验和风电机组振动信号降噪实例,证实了该方法具有良好的非线性降噪性能。

Aiming at the problem that the actual engineering vibration signals are interfered by strong noise with strong nonlinear characteristic,a phase space reconstruction method based on adaptive intrinsic dimension estimation manifold learning was proposed.Firstly,one-dimensional time series containing noise were reconstructed into a high dimensional phase space with the phase space reconstruction method.Secondly,the intrinsic dimension of each sample point in the phase space was estimated based on the maximum likelihood estimate （MLE）,the adaptive weighted average method was used to calculate the global intrinsic dimension.At last,the manifold learning algorithm and the local tangent space alignment （LTSA）were employed to project the signal containing noise from the high-dimensional phase space into the intrinsic dimensional space of useful signals.After eliminating the noise distributing in the high-dimensional space, the signals were reconstructed back into one-dimensional time series.Lorenz simulation and an example of vibration signals＇noise reduction for a wind power generator unit showed that the proposed method has a good performance of nonlinear noise reduction.