基于一种小波核优化学习的KSPP子空间故障特征提取

Method for feature extraction in KSPP feature subspace based on wavelet kernel learning

针对电子系统故障诊断中有效特征提取困难、核属性约简方法中核函数与核参数选择繁琐等问题,提出了一种基于自优化小波核稀疏保持投影的子空间特征提取方法。通过对核极化准则的改进,使得新准则不仅可以处理多类别信息,而且可以保留同一类别数据间的局部结构特征。以墨西哥帽小波核函数为对象,基于改进的核评估准则构建优化目标函数,并采用粒子群优化算法进行核参数选择;将优化的小波核作为核稀疏保持投影的核函数,最终实现了在核子空间中对有效特征的提取。实验结果表明,相比于其他流行的子空间特征提取方法,提出的方法有效提升了分类精度,具有良好的泛化性能。

In the fault diagnosis of electronic system,it is difficult to extract effectively fault features. As a result,this paper presented a new feature extraction method based on self-optimization wavelet kernel sparsity preserving projection. At first, the kernel polarization criterion was extended to an improved form so that it could simultaneously encode the muhiclass information and preserve the local structure of within-class data. For Mexico-hat wavelet kernel function, this paper established a new objective function based on improved kernel evaluation measurement criterion. Then it obtained the optimal kernel parameter by minimizing objective function based on particle swarm optimization algorithm. Finally, it extracted effective features from kernel feature subspace by inserting optimized wavelet kernel function into kernel sparsity preserving projection. Compared with several well-known feature extraction methods, experimental results show that the proposed method can obtain higher classification accuracy and better generalization performance.