摘要： 针对风电变流器故障信号非平稳、非线性的特性,结合经验模态分解(EMD)对非线性信号处理的自适应性和分形盒维数能对非线性行为定量描述的特点,提出基于EMD与分形的变流器故障特征提取方法。逆变输出三相电压信号经EMD处理后,将所得的含故障特征的固有模态分
量的信息熵作为能量分布特征,分形盒维数作为结构特征,2种特征量结合较之前单一特征量更能精确反映变流器故障状态。但该特征提取法容易引发维数灾难,因此引入有监督增量式正交判别邻域保持嵌入流形学习方法来对故障特征进行维数约简,研究参数k,d的选择问题,
加入类标签信息增强局部类内几何关系、最大化类间距离,并根据流形采样密度和曲率对k进行自适应调节。基于关联维数对吸引子不均匀性反应敏感,更能反映吸引子动态结构的特性,利用其对d进行估计,弥补通常情况下参数d难以确定的不足。通过Matlab仿真,验证了所
提方法对变流器故障识别的准确性与有效性,且识别率提高明显。
关键词:
风电变流器,
特征提取,
维数约简,
经验模态分解,
分形盒维数,
流形学习
Abstract: In order to overcome non-stationary,non-linear characteristics for wind power converter fault signal,combined Empirical Mode Decomposition(EMD) nonlinear adaptive signal processing and fractal box dimension which can quantitatively describe the nonlinear behavior,this paper proposes a new method based on EMD and fractal theory for wind power converter fault diagnosis.That is to say,firstly process the converter output three-phase voltage signal with EMD,and then act the resulting intrinsic mode components entropy contained fault features as energy distribution.Conduct fractal box dimension of each component of the IMF as structural feature.Compared with the previous single feature,the energy characteristics and structural features can act well as converter fault feature.But the method for feature extraction easily leads to curse of dimensionality.Therefore,introduce the supervised incremental orthogonal discriminant
neighborhood preserving embedding manifold learning approach to reduce the dimension of fault features and focus on the problem of how to select the parameter k and d:introduce class label information to enhance the geometric relationships within the local class,maximize the distance between classes,and adjust the parameter k adaptively according to the manifold sampling density and curvature.Correlation dimension is sensitivity to the inhomogeneity of attractor,it can better reflect the characteristics of the dynamic structure of the attractor and can be used to estimate the value of d,up to breakthrough the difficult for determining the parameters of d.By Matlab simulation,it verifies that the method is accurate and efficient to identify the type of converter fault,and the fault identification rate is greatly improved.
Key words:
wind power converter,
feature extraction,
dimension reduction,
Empirical Mode Decomposition(EMD),
fractal box dimension,
manifold learning
中图分类号: